TSW03232ESES-SWGdefinedStorage IBM PDF

Compliments of puting Edition
IBM Platform Com

Reduce storage costs using
fined
Open the book and find:
De
any hardware with storage
controlled in software
Traditional storage systems have become
• Software defined storage
systems that meet your
business needs Softwa re
Storage
costly performance bottlenecks for enterprises
struggling with ever-growing data challenges. • Performance bottlenecks
that exist in your storage
This book explains how software defined storage
infrastructure
enables organizations to significantly reduce
their storage costs while improving performance, • Features and capabilities
reliability, and scalability with any hardware of IBM General Parallel
and intelligent software that performs essential File System (GPFS)
storage functions. • Turnkey software defined
storage solutions that are
• Increase flexibility — traditional storage ready to deploy
systems limit your options and lock you • Innovative IBM GPFS use
in to a rigid and unadaptable solution cases for complex storage
• Simplify management — automated challenges in different
policy-driven storage management makes industries
it easy to implement policies for information
life cycle management and other storage
administration tasks
Making Everything Easier!™ Learn to:
• Empower global collaboration — low
Go to Dummies.com® • Control storage costs
for videos, step-by-step examples,
latency access to data from anywhere how-to articles, or to shop! • Eliminate storage bottlenecks
in the world to enable innovation and
increase productivity • Use IBM GPFS to solve storage
management challenges
Lawrence C. Miller
CISSP
ISBN: 978-1-118-84178-5
Not for resale Scott Fadden
These materials are © 2014 John Wiley & Sons, Inc. Any dissemination, distribution, or unauthorized use is strictly prohibited.
Sof tware Def ined
Storage
IBM Platform Computing Edition
by Lawrence C. Miller, CISSP

and
Scott Fadden
Software Defined Storage For Dummies®, IBM Platform Computing Edition
Published by
John Wiley & Sons, Inc.
111 River St.
Hoboken, NJ 07030-5774
www.wiley.com
Copyright © 2014 by John Wiley & Sons, Inc., Hoboken, New Jersey
No part of this publication may be reproduced, stored in a retrieval system or transmitted in any
form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise,
except as permitted under Sections 107 or 108 of the 1976 United States Copyright Act, without the
prior written permission of the Publisher. Requests to the Publisher for permission should be
addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ
07030, (201) 748-6011, fax (201) 748-6008, or online at http://www.wiley.com/go/permissions.
Trademarks: Wiley, For Dummies, the Dummies Man logo, The Dummies Way, Dummies.com,
Making Everything Easier, and related trade dress are trademarks or registered trademarks of
John Wiley & Sons, Inc. and/or its affiliates in the United States and other countries, and may not
be used without written permission. IBM, the IBM logo, and GPFS are trademarks or registered
trademarks of International Business Machines Corporation. All other trademarks are the property
of their respective owners. John Wiley & Sons, Inc., is not associated with any product or vendor
mentioned in this book.
LIMIT OF LIABILITY/DISCLAIMER OF WARRANTY: THE PUBLISHER AND THE AUTHOR MAKE

NO REPRESENTATIONS OR WARRANTIES WITH RESPECT TO THE ACCURACY OR
COMPLETENESS OF THE CONTENTS OF THIS WORK AND SPECIFICALLY DISCLAIM ALL
WARRANTIES, INCLUDING WITHOUT LIMITATION WARRANTIES OF FITNESS FOR A
PARTICULAR PURPOSE. NO WARRANTY MAY BE CREATED OR EXTENDED BY SALES OR
PROMOTIONAL MATERIALS. THE ADVICE AND STRATEGIES CONTAINED HEREIN MAY NOT BE
SUITABLE FOR EVERY SITUATION. THIS WORK IS SOLD WITH THE UNDERSTANDING THAT
THE PUBLISHER IS NOT ENGAGED IN RENDERING LEGAL, ACCOUNTING, OR OTHER
PROFESSIONAL SERVICES. IF PROFESSIONAL ASSISTANCE IS REQUIRED, THE SERVICES OF A
COMPETENT PROFESSIONAL PERSON SHOULD BE SOUGHT. NEITHER THE PUBLISHER NOR
THE AUTHOR SHALL BE LIABLE FOR DAMAGES ARISING HEREFROM. THE FACT THAT AN
ORGANIZATION OR WEBSITE IS REFERRED TO IN THIS WORK AS A CITATION AND/OR A
POTENTIAL SOURCE OF FURTHER INFORMATION DOES NOT MEAN THAT THE AUTHOR OR
THE PUBLISHER ENDORSES THE INFORMATION THE ORGANIZATION OR WEBSITE MAY
PROVIDE OR RECOMMENDATIONS IT MAY MAKE. FURTHER, READERS SHOULD BE AWARE
THAT INTERNET WEBSITES LISTED IN THIS WORK MAY HAVE CHANGED OR DISAPPEARED
BETWEEN WHEN THIS WORK WAS WRITTEN AND WHEN IT IS READ.
For general information on our other products and services, or how to create a custom For
Dummies book for your business or organization, please contact our Business Development
Department in the U.S. at 877-409-4177, contact info@dummies.biz, or visit www.wiley.com/go/
custompub. For information about licensing the For Dummies brand for products or services, con-
tact BrandedRights&Licenses@Wiley.com.
ISBN: 978-1-118-84178-5 (pbk); ISBN: 978-1-118-84314-7 (ebk)
Manufactured in the United States of America
10 9 8 7 6 5 4 3 2 1
Table of Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Chapter 1: Storage 101 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Data Access and Management Challenges.............................. 4
Three Important Functions of Storage..................................... 6
Defining Types of Storage.......................................................... 7
Hard Disk and SSD Technologies............................................ 13
Cluster File Systems................................................................. 15
Chapter 2: What Is Software Defined Storage?. . . . . . . 17

Defining Software Defined Storage......................................... 17
Key Benefits of Software Defined Storage.............................. 18
Supporting Files, Blocks, and Objects.................................... 21
Chapter 3: Digging Deeper into IBM GPFS. . . . . . . . . . . 23

Understanding GPFS Concepts............................................... 23
Removing Data Related Bottlenecks...................................... 24
Simplifying Data Management................................................. 26
Cluster Configurations............................................................. 28
Sharing Data Across GPFS Clusters........................................ 31
Managing the Full Data Life Cycle at Lower Costs................ 35
GPFS-powered Enterprise Solutions....................................... 36
Chapter 4: Getting to Know the IBM GPFS

Storage Server. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Delivering an End-to-End Solution.......................................... 40
Bringing Parallel Performance to Fault Tolerance............... 42
Bringing It All Together............................................................ 44
Chapter 5: Software Defined Storage in the

Real World. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Higher Education...................................................................... 45
Energy......................................................................................... 47
Engineering Analysis................................................................ 50
Life Sciences.............................................................................. 53
Media and Entertainment........................................................ 55
Research and Development..................................................... 57
Chapter 6: Ten Ways to Use Software Defined

Storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Publisher’s Acknowledgments
We’re proud of this book and of the people who worked on it. For details on how
to create a custom For Dummies book for your business or organization, contact
info@dummies.biz or visit www.wiley.com/go/custompub. For details on
licensing the For Dummies brand for products or services, contact
BrandedRights&Licenses@Wiley.com.
Some of the people who helped bring this book to market include the following:
Project Editor: Carrie A. Johnson

Acquisitions Editor: Connie Santisteban
Editorial Manager: Rev Mengle
Business Development Representative: Christiane Cormier
Custom Publishing Project Specialist: Michael Sullivan
Introduction
T he rapid, accelerating growth of data, transactions, and
digitally aware devices is straining today’s IT infrastruc-
ture and operations. At the same time, storage costs are
increasing and user expectations and cost pressures are
rising. This staggering growth of data has led to the need for
high-performance streaming, data access, and collaborative
data sharing.
If you work with big data in the cloud or deal with structured
and unstructured data for analytics, you need software defined
storage. Software defined storage uses standard compute,
network, and storage hardware; the storage functions are all
done in software, such as IBM GPFS, that provides automated,
policy driven, application aware storage services, through
orchestration of the underlining storage infrastructure in
support of an overall software defined environment.
IBM General Parallel File System (GPFS) provides online

storage management, scalable access, and integrated
information lifecycle management tools that are capable
of managing petabytes of data and billions of files. This
high-performance, shared-disk file management solution
offers fast, reliable access to a common set of file-based data.
About This Book

Software Defined Storage For Dummies, IBM Platform Computing
Edition, examines data storage and m anagement challenges
and explains software defined storage, an innovative solution
for high-performance, cost-effective s torage using IBM’s GPFS.
2 Software Defined Storage For Dummies
Foolish Assumptions
It’s been said that most assumptions have outlived their
uselessness, but I’ll assume a few things nonetheless! Basically,
I assume that you know a little something about storage
technologies and storage management challenges. As such, this
book is written primarily for technical readers and decision
makers, such as storage administrators and IT managers.
Icons Used in This Book

Throughout this book, you’ll occasionally see special icons
to call attention to important information. No smiley faces
winking at you or any other cute little emoticons, but you’ll
definitely want to take note! Here’s what you can expect.
This icon points out information that may well be worth

ommitting to your nonvolatile memory, your gray matter, or
c
your noggin — along with anniversaries and birthdays.
You won’t find a map of the human genome or the blueprints

for IBM’s Watson here (or maybe you will, hmm), but if you
seek to attain the seventh level of NERD-vana, perk up. This
icon explains the jargon beneath the jargon and is the stuff
legends — well, nerds — are made of.
Thank you for reading, hope you enjoy the book, please take
care of your writers. Seriously, this icon points out helpful
suggestions and useful nuggets of information.
Proceed at your own risk . . . well, okay — it’s actually nothing
that hazardous. These helpful alerts offer practical advice to
help you avoid making potentially costly mistakes.
Chapter 1
Storage 101
In This Chapter
▶ Recognizing data access and management challenges
▶ Knowing the basics of what storage does
▶ Understanding different types of storage
▶ Distinguishing between different storage technologies
▶ Looking at cluster file systems
E nterprise users and data consumers are churning out vast

amounts of documents and rich media (such as images,
audio, and video). Managing the volume and complexity of
this information is a significant challenge in organizations of
all types and sizes as more and more applications feed on and
increase this deluge of file-based content, and as individuals
and businesses collaborate for intelligence, analytics, and
information sharing.
This chapter provides some background on digital storage

technologies and the storage industry. We start by quickly
summing up storage industry challenges, and then you take a
look at what function storage serves in your IT infrastructure.
Next, you look at the hardware nuts and bolts of storing digital
data (don’t worry, it’s not too geeky) and common software
approaches to organizing data. You finish up with a look at the
beginnings of software defined storage — scalable file system
storage.
Data Access and Management

Challenges
Digital data — both structured and unstructured — is every-
where and continues to grow at a stunning pace. Every day,
approximately 15 petabytes of new information is generated
worldwide, and the total amount of digital data doubles
approximately every two years. At the same time, storage
budgets are increasing only one to five percent annually, thus
the gap between data growth and storage spending is widen-
ing (see Figure 1-1). The data growth explosion, as well as the
nature and increasing uses of data, are creating tremendous
data storage challenges for enterprises and IT departments
everywhere. Simply put, storage needs to be less expensive in
order to keep up with demand.
Structured data refers to data that’s organized, for example, in

a database. Unstructured data refers to data that doesn’t have a
defined model or framework — for example, multimedia files.
Zettabytes
Data doubles approximately
1000 X
every 2 years
Exabytes
1000 X
Petabytes
1000 X
Terabytes
Storage budgets increase only 1-5%
1000 X
Gigabytes
1980 1990 2000 2010 2013
Figure 1-1: S
torage requirements are devouring CAPEX and OPEX
resources.
Chapter 1: Storage 101 5
Some of this growth can be addressed with larger hard disk
drives and networking components getting faster and faster.
But as these technologies advance, making the data useful
becomes more difficult. Larger hard disk drives enable you to
store more data, but in many cases the hardware appliances
that utilize these drives aren’t able to keep up.
To address this problem, many IT managers today are adding

more and more network-attached storage (NAS) devices. NAS
devices can be relatively cost effective to purchase, but grow-
ing your file storage in this manner causes data administra-
tion and management (such as migration, backups, archiving)
costs to skyrocket! And while this approach may solve your
storage capacity problem, it doesn’t necessarily improve
application I/O performance and it does nothing to reduce
management costs or improve application workflow.
Many data centers have become victims of “filer-sprawl” — IT

departments deploying numerous, relatively inexpensive NAS
appliances or “filers” in a futile effort to keep up with out-of-
control storage capacity demands. Beyond simply trying to
keep up with capacity demands, “stop-gap” or temporary fixes
often lead to other data access and management challenges,
including
✓ Rising administrative costs to manage file-based data

✓ Data accessibility that’s limited in remote locations
✓ Continuous data availability and protection that becomes
increasingly difficult to maintain
✓ Backup and archival operations that can’t keep pace with
growing data
Data access and management is critical to an efficient com-

puting infrastructure. An efficient infrastructure must be
balanced properly between three key components: compute,
network, and data. The network and the data are normally the
most difficult challenges for enterprise IT departments.
Three Important Functions

of Storage
At its most basic level, enterprise storage performs three
important functions, which are
✓ Store data (intermediate and final)

✓ Protect data from being lost
✓ Feed data to the computer’s processors (so they can
keep doing work)
Storage administrators have always recognized the need for

storage capacity and data protection, and most storage ven-
dors do a good job of providing solutions that satisfy these
first two functions.
However, storage administrators are now increasingly focused

on the third function of storage because getting data from
the storage to the processor has become a performance
bottleneck, and most storage vendors have done a poor job
of addressing this performance issue. Consider that over the
past decade,
✓ CPU speed performance has increased 8 to 10 times.

✓ DRAM speed performance has increased 7 to 9 times.
✓ Network speed performance has increased 100 times.
✓ Bus speed performance has increased 20 times.
✓ Hard disk drive (HDD) speed performance has increased
only 1.2 times.
One result of this storage bottleneck is that your applications

may be running slow, which negatively impacts productivity
and wastes the capacity of other expensive infrastructure in
your datacenter.
Defining Types of Storage

Not all storage is created equal. Storage systems are
commonly divided into block-, file-, and object-based storage
systems and include direct-attached storage (DAS), network-
attached storage (NAS), and storage area networks (SAN).
Block storage
Block-based storage stores data on a hard disk as a sequence
of bits or bytes of a fixed size or length (a block). In a block-
based storage system, a server’s operating system (OS)
connects to the hard drives. Block storage is accessible via
a number of client interfaces including
✓ Fibre Channel (or Fibre Channel Protocol, FCP)

✓ SCSI (Small Computer System Interface) and iSCSI
(Internet Protocol SCSI)
✓ SAS (Serial Attached SCSI)
✓ ATA (Advanced Technology Attachment) and SATA
(Serial ATA)
Fibre Channel and iSCSI interfaces, as well as SAS, are

commonly used for storage housed outside of the computer,
such as a storage area network (SAN). SAS and ATA are
typically used in direct-attached storage (DAS).
Block-based storage systems are commonly implemented as

either direct-attached storage (DAS) or storage area networks
(SAN).
Direct-attached storage
DAS is the simplest and cheapest type of storage for computer
systems. As the name implies, DAS is directly attached to the
computer or server via a bus interface (see Figure 1-2).
Computer
Bus
Disk(s) &
filesystem
Figure 1-2: D
AS connects hard disks directly to the computer or server via
a bus interface.
DAS can provide basic data protection functions (for example,

RAID and backup) and has very limited capacity because you
can only install as many drives as the number of physical
slots available on the server.
Storage area networks

A storage area network (SAN) is a separate storage system
with its own protection functions connected to a server or
servers through a dedicated storage network (see Figure 1-3).
Server 1 Server 2 Server 3
Fibre Channel
Switch
SAN storage 2
system 1
protects and
manages data
3
Figure 1-3: A SAN connects a storage array to servers via a dedicated

storage network.
A SAN can be used by multiple servers. Each server has
one or more fast, dedicated storage connections to one or
more storage arrays. A SAN allows multiple computers to
share access to a set of storage controllers. This provides
great flexibility for maintaining enterprise IT infrastruc-
tures. In large organizations, SANs enable a division of
labor where the system administrators manage the com-
puters and the storage administrators manage the SAN.
On its own, data can’t be shared between separate LUNs or
volumes, even within the SAN, except when cluster file sys-
tems are used in the SAN. Having multiple computers sharing
access to the same data is important to many applications
and workflows — making a cluster file system a necessary
addition to SANs used for these purposes.
A LUN (Logical Unit Number) is an identifier assigned to a

collection of disks within the storage system, defined in a
storage controller and partitioned so that host servers can
access them. A computer can then use these LUNS to store
data. For example, you can create a file system on a LUN as a
place to store files. A volume is part of a LUN created within
volume management software.
The IBM XIV Storage System is an example of block-based

SAN attached storage. Learn more at www.03.ibm.com/
systems/storage/disk/xiv.
SANs are commonly used in mission-critical or high-transaction

(IOPS, or I/O Operations Per Second) environments, for
example, online transaction processing (OLTP) databases,
ERP (Enterprise Resource Planning), and virtualized systems.
Advantages of SANs include
✓ Fast speeds: SAN speeds are increasing dramatically due to

• Fabric interconnects: Speeds of 40 Gbps and
80 Gbps are common, and InfiniBand EDR (Enhanced
Data Rate) in a 12X cluster is capable of 300 Gbps
data rates.
• More spindles, more speed: As you add more drives
to a SAN, you can increase the read/write access
speeds available to the computers using the SAN.
✓ Management: Processing and storage are managed

separately.
✓ Data protection: Data protection functions, such as
backup and off-site replication, can be done outside the
computer running the application and don’t choke the
performance of the attached servers.
Compared to other storage systems, SANs can be relatively

expensive because they’re engineered for maximum reliability
and performance.
File storage
File-based storage systems, such as NAS appliances, are often
referred to as “filers” and store data on a hard disk as files in a
directory structure. These devices have their own processors
and OS and are accessed by using a standard protocol over a
TCP/IP network. Common protocols include
✓ SMB (Server Message Block) or CIFS (Common Internet

File System): SMB (or CIFS) is commonly used in
Windows-based networks.
✓ NFS (Network File System): NFS is common in Unix- and
Linux-based networks.
✓ HTTP (Hypertext Transfer Protocol): HTTP is the protocol
you most commonly use when using a web browser.
NAS appliances are relatively easy to deploy, and client access

is straightforward using the common protocols. Computers
and the NAS appliances are all connected over a shared
TCP/IP network, and the data stored on NAS appliances can
be accessed by virtually any computer, regardless of the
computer’s OS.
NAS appliances are fairly common in datacenters today.
However, NAS appliances have several significant disadvan-
tages. They’re typically slower than DAS or SAN and can be
storage performance bottlenecks because all data has to
go through the NAS’s own processors. NAS appliances also
have limited scalability. When a NAS appliance fills up, you
add another, and another, and so on. This creates “islands of
storage” that are very inefficient to manage (see Figure 1-4).
Filer 1 Filer 2 Filer 3 Filer 4 Filer 5 Filer 6 Filer 7 Filer 8
Storage Storage Storage Storage Storage Storage Storage Storage

Island Island Island Island Island Island Island Island
Figure 1-4: F ilers often lead to “islands of storage” in the datacenter.
Object storage
Object-based storage systems use containers to store data
known as objects in a flat address space instead of the hierar-
chical, directory-based file systems that are common in block-
and file-based storage systems (see Figure 1-5).
File System Object

Root Directory
c:\
Unique Object ID 83568qw4590pxvbn
DOCS
APPS
DOS
Date, size, camera Metadata
Files SPREAD GAMES File, image, video Data
/c/apps/games/b 83568qw4590pxvbn
Figure 1-5: C
omparing the file system to the object-based storage system.
A container stores the actual data (for example, an image

or video), the metadata (for example, date, size, camera
type), and a unique Object ID. The Object ID is stored in a
database or application and is used to reference objects in
one or more containers. The data in an object-based storage
system is typically accessed using HTTP using a web browser
or directly through an API like REST (representational state
transfer). The flat address space in an object-based storage
system enables simplicity and massive scalability, but the
data in these systems typically can’t be modified (other than
being completely deleted and an entirely new version written
in its place — an important distinction to keep in mind).
Object-based storage is commonly used for cloud services

by providers such as IBM SoftLayer, Amazon S3, Google, and
Facebook.
Time to explain RAID

RAID (Redundant Array of performance is slower than RAID
Independent Disks, originally 0 because data must be simulta-
Redundant Array of Inexpensive neously written to both drives in
Disks) is a data storage technology a mirrored set and usable stor-
that distributes data across multiple age capacity is reduced by 50
drives in one of several ways (called percent; one drive in a mirrored
RAID levels), depending on the set can fail without loss of data.
level of performance and protec-
✓ RAID 2 (bit-level striping with
tion required. Eight standard RAID
dedicated Hamming-code
levels are defined by the Storage
parity). Requires a minimum
Networking Industry Association
of three hard drives with each
(SNIA) as follows:
sequential bit of data striped
✓ RAID 0 (block-level striping with- across a different drive; this is
out parity or mirroring). Requires a theoretical RAID level that has
a minimum of two hard drives; not been implemented.
provides maximum performance
✓ RAID 3 (byte-level striping with
and usable storage capacity, but
dedicated parity). Requires a
no redundancy.
minimum of three hard drives
✓ RAID 1 (mirroring without parity with each sequential byte of data
or striping). Requires a minimum striped across a different drive;
of two hard drives; read perfor- not commonly used.
mance is not impacted. Write
✓ RAID 4 (block-level striping with in use today, but with growing

dedicated parity). Requires a drive capacities (in excess of 3
minimum of three hard drives; TB) rebuilds can take days with
similar to RAID 5 but with parity the entire system being very
data stored on a single drive. unresponsive during that time.
✓ RAID 5 (block-level striping with ✓ RAID 10 (mirroring and striping,
distributed parity). Requires a also known as RAID 1+0).
minimum of three hard drives; Requires a minimum of four hard
data and parity are striped drives; data is striped across
across all drives; a single drive primary disks and mirrored to
failure causes all subsequent secondary disks in an array. Read
reads to be calculated from the performance is not impacted, but
parity information distributed write performance is degraded
across the remaining functional similar to RAID 1 and usable stor-
drives, until the faulty drive is age capacity is reduced by 50
replaced and the data rebuilt percent; one drive in each span
from the distributed parity (primary and secondary) can fail
information. without loss of data.
✓ RAID 6 (block-level striping In Chapter 3, you find out how new
with double distributed parity). RAID implementations are providing
Requires a minimum of four innovative approaches to these stan-
hard drives; similar to RAID 5 but dard RAID definitions. One of these
allows for two failed drives. This new technologies is GPFS Native
is the most common RAID level RAID.
Hard Disk and SSD Technologies

The most common hard drives in use today include SATA
(Serial ATA), SAS (Serial Attached SCSI), and SSD (Solid State
Drives). Each drive technology provides different combina-
tions of capacity, performance, and cost (see Figure 1-6).
SATA drives are typically used in desktop and laptop com

puters, as well as DAS, NAS, and SAN. SATA drives provide the
highest capacity (for example, 2 to 4TB) and lowest cost per
gigabyte. However, SATA drives are slower (typically 7,200
RPM) and less reliable than other drive technologies. SATA is
commonly implemented in SANs for secondary storage and
for application data with relatively low IOPS requirements.
200 and 400GB

2.5 inch SSD
Performance (IOPS)*
$
300GB
15,000 pm
2.5 inch drives
600 and 900GB
10,000 rpm
2.5 inch drives
3 and 4TB
7,200 pm
3.5 inch drives
Capacity
Figure 1-6: D
ifferent hard drive technologies require a tradeoff between
capacity, performance, and cost.
SAS drives are commonly used in servers (DAS) or SANs. SAS

drives provide a tradeoff between performance (typically
10,000 and 15,000 RPM) and capacity (for example, 300, 600,
and 900GB). SAS drives are more reliable than SATA drives and
their individual components are designed to handle frequent
read/writes and high IOPS.
SSDs use flash technology to provide reliable and high-speed

data storage. Flash technology uses floating gate transistors
to store data as 1s and 0s in individual cells. SSD capacities
are increasing rapidly, with current capacities ranging up
to 500GB, and are extremely fast: Read/write operations on
Flash storage are measured in microseconds compared to
milliseconds for hard disk drives, and IOPS are measured
in tens of thousands to millions, compared to hundreds for
hard disk drives. Although the cost of SSDs is dropping and
capacity increasing, this technology still comes at a premium
and is most commonly used today in situations where high
performance is needed over capacity.
Cluster File Systems

A cluster file system can be accessed from many computers
at the same time over a network or a SAN. Yes, this sounds
similar to network protocols such as CIFS or NFS, but there
are a few key differences:
✓ Some cluster file systems provide access from multiple

nodes over a SAN; this isn’t possible with CIFS or NFS.
✓ Direct access from the computer using a SAN provides a
cluster file system several performance advantages over
standard network protocols.
✓ Cluster file systems are tightly coupled and communicate
at a more sophisticated level to enable an application, for
example, to have multiple nodes reading and writing to a
single file.
Some cluster file systems extend the same functionality over

TCP/IP or Infiniband. This is similar to NFS and CIFS, because
both approaches use the network to access data, but in the
case of a cluster file system, the network protocol used to
transfer data is part of the cluster file system software. This
tight integration allows the cluster file system to provide high
performance and advanced access patterns over the network.
These protocols can leverage technologies including Remote
Direct Memory Access (RDMA) for faster processing of data.
These integrations mean that cluster file system can be fast

and provide advanced functionality, but they aren’t p articularly
well suited for workstation access. You wouldn’t run a cluster
file system on your tablet to access your music, for example.
Cluster file systems are designed to provide enhanced file data
access for the IT infrastructure, like the systems from which
you download your music.
A cluster file system achieves high I/O performance by

spreading data across multiple storage servers, all sharing the
same global namespace, to increase scalability.
A parallel file system is a type of cluster file system that reads

and writes data in parallel across multiple storage nodes
providing extremely high performance, scalability, and data
protection. You find out more about parallel file systems in
Chapter 2.
Chapter 2
What Is Software Defined

Storage?
In This Chapter
▶ Putting the all the pieces together in software defined storage
▶ Recognizing the benefits of software defined storage
▶ Introducing GPFS software defined storage
S oftware defined storage is a relatively new concept in

the computing and storage industry and can refer to
many different technologies and implementations. Software
defined storage is part of a larger industry trend that includes
software defined networking (SDN) and software defined data
centers (SDDC). This chapter explains exactly what software
defined storage is.
Defining Software Defined

Storage
At its most basic level, software defined storage is enterprise
class storage that uses standard hardware with all the
important storage and management functions performed
in intelligent software. Software defined storage delivers
automated, policy-driven, application-aware storage
services through orchestration of the underlining storage
infrastructure in support of an overall software defined
environment. Standard hardware includes
✓ Disk storage such as SAN, NAS, and disk arrays or JBODs

(just a bunch of disks)
✓ Network devices such as switches and network
interfaces
✓ Servers for storage processing, management, and
administration
Additional characteristics of software defined storage

can include
✓ Automated policy-driven administration for storage

management functions, such as information life cycle
management (ILM) and provisioning
✓ Storage virtualization
✓ Separate control and data planes to manage the storage
infrastructure and data in the storage infrastructure,
respectively
✓ Massive scale-out architecture
These characteristics are in contrast to traditional storage

systems that depend heavily on custom hardware-based
controllers to perform storage functions. NAS, DAS, and SAN
(discussed in Chapter 1) are examples of typical hardware-
based storage systems that rely on special RAID controllers
and non-portable custom firmware for their storage functions.
Key Benefits of Software

Defined Storage
Enterprises today are recognizing many significant benefits of
software defined storage in their datacenters. These include
increased flexibility, automated management, cost efficiency,
and limitless scalability.
Increased flexibility and agility

Traditional enterprise storage platforms such as SAN and NAS
(discussed in Chapter 1) are typically based on proprietary
systems and come with a high total cost of ownership (TCO). SAN
solutions typically require the use of costly and complex SAN
switches, storage arrays, and other proprietary components.
Chapter 2: What Is Software Defined Storage? 19
NAS devices are relatively inexpensive but have limited
scalability. When you run out of space on a NAS, you simply
add more NAS devices. However, this isn’t a true scale-out
capability because each individual NAS is presented as
separate, standalone storage that’s separately managed.
A software defined storage solution increases flexibility by

enabling organizations to use non-proprietary standard
hardware and, in many cases, leverage existing storage
infrastructure as part of their enterprise storage solution.
Additionally, organizations can achieve massive scale with
a software defined storage solution by adding individual,
heterogeneous hardware components as needed to increase
capacity, and improve performance in the solution.
Intelligent resource utilization

and automated management
Automated, policy-driven management of software defined
storage solutions helps drive cost and operational efficiencies.
As an example, software defined storage manages important
storage functions including ILM, disk caching, snapshots,
replication, striping, and clustering. In a nutshell, these
software defined storage capabilities enable you to put the
right data in the right place, at the right time, with the right
performance, and at the right cost — automatically.
Cost efficiency
Rather than using expensive proprietary hardware, software
defined storage uses standard hardware to dramatically lower
both acquisition costs and TCO for an enterprise-class storage
solution. The software in a software defined storage solution is
standards based and manages the storage infrastructure as well
as the data in the storage system.
In many cases, organizations can leverage their existing

investments in storage, networking, and server infrastructure
to implement an extremely cost-effective software defined
storage solution.
In a July 2012 report, Gartner, Inc., found that the average

acquisition cost per gigabyte of traditional multi-tiered storage
systems ranged from $0.9/GB to $5/GB. By comparison, software
defined storage solutions average $0.4/GB.
Limitless elastic data scaling

Unlike traditional storage systems, such as SAN and NAS,
software defined storage enables you to scale out with relatively
inexpensive standard hardware, while continuing to manage
storage as a single enterprise-class storage system. As you scale
out your storage infrastructure, performance and reliability
continue to improve. As an example, IBM General Parallel
File System (GPFS) delivers orders of magnitude more in I/O
performance improvement as hardware is added, compared
to conventional NAS (see Figure 2-1). GPFS is covered in more
detail in Chapter 3.
Measured I/O
performance
IBM GPFS
Conventional NAS
Figure 2-1: I BM GPFS delivers extreme I/O performance.
Software defined storage provides massive, virtually limitless

scalability. For example, IBM GPFS supports
✓ A maximum file system size of one million yottabytes

✓ 263 (or approximately 9 quintillion) files per file system
✓ IPv6
✓ 1 to 16,384 nodes in a cluster
A yottabyte is equal to one trillion terabytes.
Chapter 2: What Is Software Defined Storage? 21
Supporting Files, Blocks,

and Objects
Software defined storage typically refers to software that man-
ages the creation, placement, protection, and retrieval of data. IBM
GPFS is an enterprise-class software defined storage solution
that runs on IBM and third-party platforms and is available as
software that can be deployed on a variety of commodity hard-
ware systems and as part of an integrated “appliance” — the
GPFS Storage Server (GSS), described in Chapter 4.
Developers of large Cloud-scale applications have been par-

ticularly interested in software defined storage. For them,
existing heavy hardware controlled storage solutions simply
don’t scale, are prohibitively expensive, and are too inflexible
to dynamically expand capacity for application data they
envision driving their business needs moving forward. Many
of them have focused their development on OpenStack — the
open source cloud computing platform for public and private
clouds. See the nearby sidebar “Focusing on OpenStack” for
more information.
For OpenStack developers, GPFS can unify your storage with

a common way to store VM images, block devices, objects, and
files. Software defined storage allows this type of integration.
GPFS functions like GPFS Native RAID (GNR) and policy-based
data placement give you the flexibility to put the data in the
best location on the best tier (performance and cost) at the
right time. Software defined storage means that you can deploy
this on heterogeneous industry standard hardware. This is
shown in Figure 2-2.
Single Name Space

OpenStack
NFS ISCSI Nova VMWare
GPFS Swift Cinder VADP SRM Hadoop
POSIX CIFS
Manila Glance VAAI vSphere
Active
Cloud
GPFS File
Manager
GPFS Storage
GPFS SSD Fast Tape

Disk GNR
Policies for Tiering, Data Distribution, Migration to Tape and Cloud
Figure 2-2: G
PFS provides a common storage plane.
Focusing on OpenStack
OpenStack has a modular archi- and OpenStack is to create a single
tecture with various components scale-out data plane for the entire
including data center or multiple connected
data centers worldwide.
✓ OpenStack Compute (Nova):
A cloud computing fabric GPFS unifies OpenStack VM images,
controller block devices, objects, and files with
support for Nova, Cinder, Swift, and
✓ Block Storage (Cinder): Provides
Glance, along with POSIX interfaces
persistent block-level storage
like NFS and CIFS for integrating
devices
legacy applications. The ability to use
✓ Object Storage (Swift): Scalable a single GPFS file system to manage
redundant storage system volumes (Cinder), images (Glance),
shared file systems (Manila), and
With OpenStack, you can control
use file clones to efficiently/quickly
pools of processing, storage, and
shared data within and between
networking resources throughout
components will be a big advantage
a datacenter. And while OpenStack
for Cloud-scale application develop-
provides open source versions of
ers using OpenStack.
block and object storage, many
OpenStack developers have identi- The robustness and features of
fied a need for more robust storage GPFS combined with OpenStack
to support Cloud-scale applications. Swift object extensions could pro-
While many OpenStack develop- vide an enterprise-grade object
ers feel they can architect around store with high storage efficiency,
limitations in OpenStack compute tape integration, wide-area replica-
capabilities and robustness, storage tion, transparent tiering, checksums,
has a much “higher bar” as far as snapshots, and ACLs — capabilities
resiliency and reliability go. most object-based storage offerings
can’t match today. OpenStack on
Responding for the need for robust

GPFS delivers compelling efficiencies
software defined storage, the
in a single unified storage solution
OpenStack “Havana” release includes
that can support object and file
an OpenStack Block Storage Cinder
access to the same data with robust
driver for IBM GPFS, giving architects
and efficient GPFS Native RAID data
who build public, private, and hybrid
protection. OpenStack Swift object
clouds access to the features and
storage on GPFS can reduce the
capabilities of the industry’s leading
amount of raw storage you need to
enterprise software-defined storage
use compared to object storage sys-
system. And the Cinder is just the
tems that rely strictly on replication.
beginning. IBM’s vision for GPFS
Chapter 3
Digging Deeper into

IBM GPFS
In This Chapter
▶ Understanding IBM GPFS
▶ Removing storage performance bottlenecks
▶ Eliminating storage management complexity
▶ Exploring different clustering options
T o really understand IBM General Parallel File System

(GPFS), you can take four servers and use a storage-area
network (SAN) to attach the storage to all four servers so
all servers can see all the disks. You install evaluation GPFS
software on the four servers and from one server create a file
system, mount that file system on all the servers, install your
application, and you’re off and running. In this chapter, you
discover the basics of the GPFS, including the core concepts,
key features, and cluster configuration options.
Understanding GPFS Concepts

GPFS started out as a clustered file system and has evolved
into so much more. Today it’s a full-featured set of file manage-
ment tools, including advanced storage virtualization, integrated
high availability, automated tiered storage management, and
performance to effectively manage very large quantities of file
data. GPFS is designed to support a wide variety of application
workloads and has been proven extremely effective in very
large, demanding environments.
GPFS allows a group of servers concurrent access to a

common set of file data over a common SAN infrastructure,
a network, or a mix of connection types. The servers can run
any mix of AIX, Linux, or Windows Server operating systems.
GPFS provides storage management, information life cycle
management tools, centralized administration, and it allows
for shared access to file systems from remote GPFS clusters
providing a global namespace.
A GPFS cluster can be a single server, two servers providing a

high-availability platform supporting a database application,
for example, or thousands of servers used for applications
such as the modeling of weather patterns. GPFS was designed
to support high-performance workloads and has since been
proven very effective for a variety of applications. Today, GPFS
is installed in clusters supporting big data analytics, gene
sequencing, digital media, and scalable file serving. These appli-
cations are used across many industries including financial,
retail, digital media, biotechnology, science, and government.
GPFS provides a unique set of extended interfaces that can

be used to provide advanced application functionality. Using
these extended interfaces, an application can determine
the storage pool placement of a file, create a file clone, and
manage quotas. These extended interfaces provide features
in addition to the standard POSIX (Portable Operating System
Interface) interfaces.
POSIX is an IEEE (Institute of Electrical and Electronics Engineers)

family of standards for maintaining compatibility between differ-
ent variations of UNIX and other operating systems.
Removing Data Related

Bottlenecks
Over the past decade, processors, memory, network, and bus
performance have all increased exponentially, but disk speed
performance has only increased 1.2 times. This performance
gap slows data heavy applications, delays schedules, and
wastes expensive infrastructure. GPFS accelerates time to
results and maximizes utilization by providing parallel access
to data. GPFS provides extreme performance and eliminates
storage bottlenecks, by providing parallel access to data.
(see Figure 3-1).
Chapter 3: Digging Deeper into IBM GPFS 25
Computer Cluster
Linux, Windows, AIX
Parallel
Access
Single name space
GPFS Storage Server Cluster

Figure 3-1: G
PFS eliminates storage bottlenecks.
GPFS achieves high-performance I/O by
✓ Striping data across multiple disks attached to multiple

servers
✓ Providing efficient client side caching
✓ Executing high-performance metadata (inode) scans
✓ Supporting a wide range of file system block sizes to
match I/O requirements
✓ Utilizing advanced algorithms that improve I/O
operations
✓ Using block-level locking based on a very sophisticated

token management system to provide data consistency,
while allowing multiple application servers concurrent
access to the files
When many servers need to use the same set of files at the
same time, the file system needs to ensure that all the files
are protected, so one server can’t change a file without the
other servers knowing about the change. Keeping thousands
of servers “in the loop” on file status is difficult and scaling up
is even harder.
GPFS provides file integrity protection through a token process

that keeps file data consistent by always ensuring there is only
one owner for any given file. This method scales well in GPFS
because any server in the cluster can be assigned file protection
duty. There are two parts to managing tokens and file consistency:
handing out the tokens and keeping file metadata up to date.
The server(s) that initially have the token for all files that are
not in use is called the token manager. You can assign one
or more servers to be a token manager. Multiple token man-
agers help each other out by sharing the workload and by
taking over when a fellow token manager fails. When a file is
opened, the token manager hands off the token for that file
to the server that’s opening the file. The server using the file
is now responsible for all metadata changes to that file. If a
server wants to open a file that is already open on another
server, the token manager redirects the request to the server
that already has the file open and lets the two servers work
out the details among themselves. This sharing of metadata
maintenance across the entire cluster is what makes GPFS scale
very effectively. Many other file storage technologies rely on a
single metadata server or centralized database. Such a single/
centralized approach quickly limits how much data can be
stored. GPFS addresses this limitation by sharing the workload.
Simplifying Data Management

Creating a billion files isn’t that difficult, but maintaining
a storage solution containing a billion files or more takes
industrial class tools. GPFS has the tools needed to manage
petabytes of data and billions of files. The global namespace
is easy to administer and can be scaled quickly, as desired, by
simply adding more scale-out resources — eliminating “filer
sprawl” and its associated issues.
GPFS has an administration model that’s easy to use and
consistent with standard file system administration practices.
These functions support cluster management and other stan-
dard file system administration functions such as user quotas
and snapshots.
A single GPFS command can perform a file system function

across the entire cluster, and most can be issued from any
server in the cluster. Optionally, you can designate a group of
administration servers that can be used to perform all cluster
administration tasks, or only authorize a single login session
to perform admin commands cluster-wide. This allows for
higher security by reducing the scope of server-to-server
administrative access.
You can set quotas by user, group, or at the directory level

to monitor and control file system usage across the cluster.
When using quotas, you can easily see usage reports that
include user, group, directory, inode, and data block usage.
You can use snapshots to protect data from human error. You
can create a snapshot of an entire file system or a subdirec-
tory (called a fileset). A snapshot is used to preserve the file
system’s contents at a single point in time. It contains a copy
of only the file system data that has changed since the last
snapshot was created and keeps that data in the same pool as
the original file, which keeps space usage at a minimum. Using
the same pool simplifies storage administration because you
don’t have to set aside additional space for snapshot data.
Snapshots provide an online backup capability that allows
you (or an end user) to easily recover from an accidental file
deletion, or the ability to compare a file to an older version.
Most application clusters need a method to get data into and

out of the cluster. Because GPFS runs directly on a standard
server, you can use a variety of tools to get file data into and
out of a GPFS file system. To better enable end user access to a
GPFS file system, the file system can be exported to clients out-
side the cluster through NFS (Network File System), including
the capability of exporting the same data from multiple servers.
This GPFS feature is called Clustered NFS (cNFS). Clustered NFS
allows you to provide scalable file service with simultaneous
access to a common set of data from multiple servers. The
cNFS feature includes failover capability, so if a NFS server fails,
the clients connected to that server automatically connect to
another server in the cluster.
NFS is a Network file system protocol that enables access to

storage by using a standard protocol over a TCP/IP network.
NFS protocol access is commonly provided by a network-
attached storage (NAS) appliance or similar device. Samba
enables file and print services for Microsoft Windows clients
from UNIX and Linux based servers.
Cluster Configurations
When it comes to cluster configuration options, GPFS is a
multi-function tool. The same GPFS software is installed on
all the servers in a cluster. What a server does, and how it
participates in the cluster, is based on the hardware it has
available and what you need it to do. Cluster configuration
is independent of which file system features you require.
Cluster configuration options can be characterized into the
three categories covered in this section.
Shared disk
A shared disk cluster is the most basic environment. In this
configuration, the storage is directly attached to all servers
in the cluster, as shown in Figure 3-2. Application data flows
over the SAN, and control information flows among the GPFS
servers in the cluster over a TCP/IP network.
GPFS Nodes
SAN
SAN Storage
Figure 3-2: S
AN attached storage.
This configuration is best for small clusters (1 to 50 servers)
when all servers in the cluster need the highest performance
access to the data. For example, this configuration is good
for high-speed data access for digital media applications or a
storage infrastructure for data analytics.
Don’t have a SAN? No problem!

When every server in the cluster can’t be attached directly
to a SAN, GPFS uses a block-level interface over the network
called the Network Shared Disk (NSD) protocol. GPFS trans-
parently handles I/O requests, whether using NSD or a direct
attachment to the SAN, so the mounted file system looks the
same to the application.
In general, SAN-attached storage provides the highest per-

formance, but the cost and management complexity of SANs
for large clusters is often prohibitive. In these cases, network
block I/O provides a viable option.
GPFS clusters use the NSD protocol to provide high-speed

data access to applications running on LAN-attached servers.
Data is served to these client servers from an NSD server. In
this configuration, disks are SAN-attached only to the NSD
servers. Each NSD server is attached to all or a portion of the
disks. Multiple NSD servers should be deployed to serve each
disk in order to avoid a single point of failure.
GPFS uses a network to transfer control information and data

to NSD clients. The network doesn’t need to be dedicated to
GPFS, but it should provide sufficient bandwidth to meet your
GPFS and other applications sharing the bandwidth.
In a NSD server configuration, a subset of the total server

population is defined as NSD servers. The NSD servers are
responsible for the abstraction of disk data blocks across an
IP-based network. The fact that I/O is remote is transparent to
the application. Figure 3-3 shows an example of a configura-
tion where a set of compute servers are connected to a set
of NSD servers via a high-speed interconnect or an IP-based
network (such as Ethernet). In this example, data to the NSD
servers flows over the SAN, and data and control information
flows to the clients across the LAN.
NSD Clients
LAN
NSD Servers
SAN
SAN Storage
Figure 3-3: N
etwork block I/O.
The choice of how many servers to configure as NSD serv-

ers is based on individual performance requirements and the
capabilities of the storage subsystem. High bandwidth LAN
connections should be used for clusters requiring significant
data transfer. To enable high-speed communication GPFS
supports TCP/IP using whatever hardware you have avail-
able (1Gbit and 10Gbit Ethernet, link aggregation, IPoIB), and
RDMA on InfiniBand for control and data communications.
InfiniBand is a switched fabric communications link com-
monly used in High Performance Computing (HPC) installa-
tions. With speeds up to 56Gbit/sec and very low latency it is
well suited to workloads with high I/O demands.
GPFS provides the ability to designate separate IP interfaces
for intra-cluster communication and the public network. This
provides a more clearly defined separation of communication
traffic. An NSD server architecture is well suited to clusters
with sufficient network bandwidth between the I/O servers and
the clients. For example, statistical applications like financial
fraud detection, supply chain management, or data mining.
Empowering global collaboration

GPFS provides low latency access to data from anywhere in the
world with Active File Management (AFM) distributed disk cach-
ing technology. AFM expands the GPFS global namespace across
geographical distances, providing fast read and write perfor-
mance with automated namespace management from anywhere
in the world. As data is written or modified at one location, all
other locations get the same data with minimal delays. These
game-changing capabilities accelerate project schedules and
improve productivity for globally distributed teams.
Sharing Data Across

GPFS Clusters
You can share data across GPFS clusters via two methods:
GPFS multi-cluster and Active File Management (AFM).
GPFS multi-cluster
GPFS multi-cluster allows you to utilize the GPFS NSD protocol
to share data across clusters. With this feature, you let other
clusters to access one or more of your file systems, and you
mount file systems that belong to other GPFS clusters for
which you’ve been authorized. A multi-cluster environment
permits the administrator access to specific file systems from
another GPFS cluster. This feature permits clusters to share
data at higher performance levels than file sharing technolo-
gies like NFS or CIFS.
GPFS multi-cluster isn’t intended to replace file sharing tech-

nologies that are optimized for desktop access or for access
across unreliable network links. Multi-cluster capability is
useful for sharing across multiple clusters within a single
physical location or across multiple locations.
In Figure 3-4, Cluster A owns the storage and manages the

file system. It may grant access to file systems that it man-
ages to remote clusters, such as Cluster B. In this example,
Cluster B doesn’t have any storage but that isn’t a require-
ment. Commonly in the case where a cluster doesn’t own
storage, the servers are grouped into clusters for ease of
management. When the remote clusters need access to the
data, they mount the file system by contacting the owning
cluster and passing required security checks. In Figure 3-4,
Cluster B accesses the data through the NSD protocol.
LAN
SAN
Cluster A Cluster B
Figure 3-4: M
ulti-cluster configuration.
Multi-cluster environments are well suited to sharing data

across clusters belonging to different organizations for collab-
orative computing, grouping of clients for administrative pur-
poses or implementing a global namespace across separate
locations. A multi-cluster configuration allows you to connect
GPFS clusters within a data center, across campus, or across
reliable WAN links.
World-wide data sharing
So what can you do if your network link is really long or not
so reliable? You can use a feature in GPFS called Active File
Management (AFM). AFM allows you to create instead of
direct access to the data in the other cluster, as in a multi-
cluster situation, a copy of the data when and where you need
to access it.
At first glance, AFM may seem like any other cache. But when
you start looking at what you can do with these basic behaviors,
the options start multiplying. To understand some of the
possible options, take a look at how AFM can be used.
A cache relationship is defined when you create a fileset. You

can have up to 1,000 independent cache relationships in each
GPFS file system. Each cache fileset can have its own mode
and configuration parameters, and still share a common set
of storage. A cache file system can be an entire file system
in size, or it can be limited by quotas. A read-only cache
contains data that can’t be modified — it’s read-only. Data is
copied into the cache fileset by GPFS, on demand, by open-
ing a file or running a command to prefetch (or pre-cache) a
list of files. A single target can have as many read-only cache
relationships as it has bandwidth available, because the cache
does all the work. A target doesn’t know there’s a cache, and
a cache only knows about its own target. Two filesets in the
same file system can be caching data from the same target,
and they wouldn’t know about each other.
The isolation between the cache and target is what makes this
model scale so well. You can have as many as 1,000 read-only
caches because each cache only has to track one relationship.
And then it gets interesting! Beyond the many-to-one cache
relationships, you can also cascade caches. A target can have
a dual personality; it can be a cache and a target at the same
time. For example, say a target exists in New York. The data
originates in London and is cached in New York. The office in
Tokyo needs a copy of the same data, so the cache in Tokyo
uses the copy in New York as the target.
In read-only mode, data consistency is simple: The target is

the only place where the data can be created and modified.
But there are other caching modes that provide additional
functionality. For example, in an independent writer cache,
data can be created and modified in the cache and at the

target. When a file is created or modified in the cache, the
write operation completes locally and the changes are auto-
matically copied to the target. You can also have multiple
independent writers updating data for the same target.
Using independent writer with multiple caches accessing the

same target is similar to having multiple people share an NFS
project space. If you copy your version of a project plan after
your colleague has saved their version of the same project
plan, the next time it’s opened, the reader sees your version.
Independent writer works in the same way, except it’s the last
cache to update that wins, so you need to think about the
file updates that could occur in a remote data center when
designing your independent writer cache relationships.
By using these basic caching techniques, you can allow all

your sites to see the same data all the time. Figure 3-5 shows
one example of using AFM to create a single view of all the
data from multiple sites.
Store1
Cache:
/data1
Clients access:
/global/data1
Local:
/global/data2 /data3
/global/data3 Store2
/global/data4
Cache: Clients access:
/global/data5 Local:
/data5 /global/data1
/global/data2
/global/data3
Cache:
/global/data4
/data3
/global/data5
Store3 /global/data6
Cache:
Clients access: /data5
/global/data1 Cache:
/global/data3
/global/data4 Cache:
/global/data6
Local:
/data5
Figure 3-5: G
lobal namespace using AFM.
In this example, each site is the target for one third of the
data. The other two sites have cache relationships with the
other sites. This means that no matter which site you log into,
you see exactly the same file system structure and you have
access to all the data — it may just take a little longer to read
a file if it has not yet been copied to your site.
Managing the Full Data Life

Cycle at Lower Costs
GPFS enhances information life cycle management and lowers
your data management costs significantly by using multiple
tiers of storage including tape. With powerful policy-driven
automation and tiered storage management, you can create
optimized tiered storage pools by grouping disks based on
performance, locality, or cost characteristics. Data migrated
to tape remains visible in the file system and is directly acces-
sible by end-users. Migration policies transparently move data
from one storage pool to another without changing the file’s
location in the directory structure.
For example, you can create a rule for thresholds that moves
files out of the high performance pool if it’s more than 80 per-
cent full, thereby mitigating potential bottlenecks in the high
performance pool. GPFS information life cycle management
capabilities and benefits include
✓ Policy-driven automation and tiered storage management

✓ Flexibility to match the cost of storage to the value of
data
✓ Storage pools create tiers of storage that include high-
performance SSD, high-speed SAS Drives, and high-
capacity NL SAS Drives
✓ Full integration with IBM Tivoli Storage Manager (TSM)
and Linear Tape File System (LTFS) that provides the
following functionality
• GPFS handles all metadata processing, then hands
the data to TSM for storage on tape
• Data is retrieved from the external storage pool
on demand when an application opens a file, for
example
• Policies move data from one pool to another with-
out changing the file’s location in the directory
structure
• Thresholds can move files out of the high-
performance pool if more than 80 percent full,
for example
Tape migration provides tiered data storage for approximately

one-fifth the cost per terabyte of disk.
GPFS-powered Enterprise
Solutions
GPFS is under the covers of some of the most game-changing
enterprise products today, including SAP HANA in-memory
database appliance, IBM DB2 PureScale, IBM SONAS (Scale-
out NAS).
SAP HANA in-memory

database appliance
SAP’s High-performance Analytic Appliance (HANA) enables
organizations to instantly explore and analyze all of their
transactional and analytical data from virtually any data
source in near real-time. Delivered on optimized hardware,
HANA realizes the efficient processing and analysis of massive
amounts of data by packaging SAP’s intelligent use of in-memory
technology, columnar database design, data compression,
and massive parallel processing together with essential tools
and functionality (for example, data replication and analytic
modeling), business content, and the SAP BusinessObjects
Business Intelligence (BI) solutions.
The three major components of this solution include IBM

Blade Server and Chassis, IBM GPFS File System, and IBM
DS4700 Storage equipment. The flexibility provided by this
solution enables customers to grow their information ware-
house as the data grows through smooth extension and
upgrade paths in a pay-as-you-grow model.
IBM DB2 pureScale

DB2 pureScale is a cluster-based, shared-disk architecture
that reduces costs through efficient use of system resources
and allows users to scale out a database on a set of servers
in an “active-active” configuration delivering high levels of
availability, scalability, and application transparency. A DB2
server that belongs to a PureScale cluster is called a member.
Each member can simultaneously access the same database
for both read and write operations. Designed mainly for OLTP
(online transaction processing) workloads with many con-
current transactions, DB2 PureScale offers almost unlimited
scale-out in terms of number of members that it can support.
Tests with up to 128 members show near-linear scalability.
GPFS is the underlying file system of IBM’s DB2 PureScale

offering for businesses with applications running on DB2 that
require on-demand storage growth and continuous availability
without any changes to the existing applications. GPFS allows
the DB2 engine to run on several host servers that cooperate
to provide coherent access to database clients from any of the
member servers. Automatic load balancing takes care of client
re-routing without the client having to reconnect to a differ-
ent server during load redistribution when a server is added,
removed, or fails.
This kind of data sharing and availability is provided by the

GPFS storage infrastructure. GPFS enables DB2 PureScale
to significantly reduce the risk and cost of business growth
and provides unlimited capacity. Pairing GPFS file systems
and DB2 for Linux, UNIX, and Windows databases creates an
unbeatable platform for high-performance database systems.
IBM SONAS
Excessive proliferation of NAS (discussed in Chapter 1)
and standalone file servers to cater to rapidly growing
enterprise storage demands has resulted in an increase in
storage system disparity and administrative and energy costs
of IT assets. Typically, more than 50 percent of user data in an
enterprise is inactive. This is because projects that once used
such data actively are no longer current, or because technolo-
gies and the businesses have evolved. Manual data life cycle
management in such cases is neither a scalable proposition
nor does it address demanding I/O-intensive application
workloads and performance.
IBM’s Scale-Out NAS (SONAS) can manage multiple

petabytes of storage and billions of files using a single file
system providing much needed operational efficiency with
automated, policy-driven tiered storage. SONAS can satisfy
bandwidth-hungry applications with its scale-out performance

and provide business continuity and disaster recovery using
asynchronous replication. SONAS leverages GPFS to provide
scalable storage and efficient data retrieval through content
metadata policies. SONAS supports storage at scale through
GPFS global namespace and hierarchical storage-management
features.
GPFS can easily scale from a two server high-availability

cluster to a 1,000 server compute farm. It is very cost effective
at any scale and provides the same set of enterprise class
functionality to a local health care clinic as it does to the
world’s largest banks. Although you may not need a file system
that supports 18 petabytes and billions of files, you will.
Inevitably, our data storage needs grow. Today, you are very
likely carrying a smartphone or USB thumb drive that has
more capacity than enterprise hard drives did just 15 years
ago — yet it still seems to never be enough!
Chapter 4
Getting to Know the IBM

GPFS Storage Server
In This Chapter
▶ Taking a look at GSS components
▶ Understanding GSS fault tolerant features
D isk drive performance hasn’t changed significantly in

many years, but in the last ten years disk drive capacity has
increased from a maximum of 100GB per disk to 4,000GB — a
40x increase! To put this in context, assume that you can read
a single disk drive at a speed of 60MB per second. In 2003,
backing up that drive took 28 minutes. Today, because disk
drives haven’t gotten much faster, backing up a single disk
full of data takes almost 19 hours. This same performance
issue also impacts disk maintenance operations, such as RAID
rebuilds. When a 4TB disk fails, how long does it take to read
and write enough data to put a new disk in its place? These
types of performance issues are one of the key driving forces
behind the invention of the IBM GPFS Storage Server (GSS).
This chapter talks about the IBM System x GSS. GSS combines
the performance of System x servers with GPFS software to
offer a high-performance, scalable building-block approach to
modern storage needs, known as software defined storage. GSS
allows you to start with a configuration that meets your organi-
zation’s current needs and expand capacity and bandwidth with
each additional GSS to meet your future needs.
Delivering an End-to-End Solution

A GPFS Storage Server is made up of IBM System x servers,
storage enclosures, and hard drives (JBODs, or just a bunch
of disks), and includes all of the software and networking com-
ponents in a fully integrated solution. IBM thoroughly tests
and optimizes each GSS for reliability, interoperability, and
performance so organizations can quickly deploy the system
and get to work achieving their business goals.
GSS is built and tested by IBM and includes solution-level

support that includes both IBM and third-party components.
Features of GSS include
✓ Scalable building-block approach to storage

✓ Two configurations: Model 24 (232 Near Line Serial
Attached SCSI, or NL/SAS Drives) and Model 26
(348 NL/SAS Drives)
✓ Up to 4TB drives
✓ Up to1392TB of capacity
✓ FDR InfiniBand, 10 and 40Gbit Ethernet interconnects,
or both
✓ Declustered RAID for sustained, predictable performance
and rapid rebuilds
✓ No hardware controllers; disk management and RAID are
performed by the GPFS Native RAID (GNR) feature
GSS packaging
GPFS Native RAID (GNR) software hardware has a specially designed
is capable of running on a variety disk tray that holds 384 drives. This
of hardware platforms, but to be hardware is excellent but not prac-
a reliable solution it needs to be tical for many customers, so IBM
tested with each server and disk released the GPFS Storage Server
tray. The GPFS native RAID soft- as a means to provide this advanced
ware was first delivered on the IBM storage technology on standard
P775 Supercomputing platform. This hardware to more customers.
Chapter 4: Getting to Know the IBM GPFS Storage Server 41
How does GSS defy physics?

Well, it doesn’t. As physics profes- In a declustered array each block of
sors will be quick to point out, you file data is cut into what is called a
can’t defy physics. But you can take strip. So if you use an 8+2 configura-
a completely new approach to solv- tion you get 8 data strips and 2 parity
ing the problem. GSS incorporates a strips the difference is that this strip-
new approach to solve the issue of ing is independent of the number of
long RAID rebuild times with really available disks. This means that if, for
large disks called declustered RAID. example, you have 58 drives and you
are using 8+2 parity, the first block of
In a traditional RAID 6 array, for
data is spread over 10 of the drives,
example, a set of disks are grouped
the second block of data is spread
together, and data is stored on all
over 10 other drives, and so on.
of the disks in the set along with
chunks of information that can be So, how does this improve rebuild
used to “rebuild” a failed drive (see speeds? Instead of having to rebuild
Chapter 1 for an explanation of RAID a failed disk as in RAID 6, a de-
6). These chunks of information are clustered RAID array just needs to
called parity bits. When a drive fails ensure that all the data that was on
in a RAID 6 array, parity bits on the the failed drive is copied or com-
other disks are read to re-construct puted (parity is computed) on other
the data from the failed drive, which drives. So GSS doesn’t defy physics,
is then written to a new disk. The it just spreads out the work. Instead
entire contents of the failed disk must of writing 4TB to a single spare drive,
be written to the new disk before the the surviving 57 drives just need to
array is again completely protected swap some data and move on.
from further disk failures.
A rebuild on a RAID 6 array with 4TB
In a declustered RAID array, the drives can negatively impact the
number of disks in a group, called performance of your entire array
a declustered array, is dynamic and for 17 to 24 hours. In a declustered
the data and parity are across the RAID array, the impact is less than
available disks as the data arrives. 15 minutes.
Figure 4-1 shows the hardware that’s used in the GSS and the
different shipping configurations.
Figure 4-1: T he IBM System x GSS.
Bringing Parallel Performance

to Fault Tolerance
The speed of recovery along with other advanced data pro-
tection features makes GSS much more reliable, even as the
number of drives increases. GSS declustered RAID technology
is not tied to a set number of drives. In GSS, there are standard
configurations, but the RAID technology can work with as few
as 11 drives to several hundred drives. This keeps GSS relevant,
even with new and different drive packages and types.
Advanced data protection

For a long time, GPFS has been able to rely on the RAID
controller to take care of things like making sure what you
write to the disk last week is correct when you read it this
week. Managing the data all the way down to the raw device
comes with additional responsibility and some great func-
tional benefits. One of these responsibilities is to make sure
the data hasn’t changed since it was written to disk. In GSS,
this is accomplished using a data checksum. A checksum is a
Chapter 4: Getting to Know the IBM GPFS Storage Server 43
number that’s computed by looking at a chunk of raw data.
This number is stored in a separate area and used when the
data is read to ensure it hasn’t changed since it was written.
Because GPFS now has that checksum information, it can be
used to ensure data integrity all the way from the disk to the
GPFS client, thereby protecting data from storage, system, or
network errors.
Another responsibility GPFS now holds is to make sure the

disks are healthy. Now all storage components are monitored
and maintained through the disk hospital. If there are media
errors, path issues, or simply disks behaving badly, GPFS
responds to these events with a series of actions, including
rerouting requests and power-cycling disks.
Hard disks don’t report some read faults and occasionally fail
to write data, while actually claiming to have written the data.
These errors are referred to as silent errors, phantom-writes,
dropped-writes, and off-track writes. GPFS Native RAID imple-
ments an end-to-end checksum calculated and appended to
the data by the client to detect silent data corruption.
If the checksum or version numbers are invalid on read, GPFS

Native RAID reconstructs the data using parity or replication,
and returns the reconstructed data and a newly generated
checksum to the client. Thus, both silent disk read errors and
lost or missing disk writes are detected and corrected.
More efficient use of

disk performance
Compared to conventional RAID, GPFS Native RAID implements
a sophisticated data layout scheme that uniformly spreads
(or “declusters”) user data, redundancy information across
all the 58 disks of a declustered array.
With declustered RAID, you don’t have idle spare disks sitting
and waiting to be called into service. Spreading data over all
available drives including spares, a hot spare becomes hot
spare space. So, instead of assigning a drive to be a spare, the
RAID software just keeps space free for failure events.
A declustered array can significantly shorten the time to

recover from a disk failure, which reduces the rebuild over-
head for client applications. When a disk fails, data is rebuilt
using all 57 operational disks in the array, thus providing

bandwidth that is 6 times greater than the 10 disks in a
conventional RAID 6 group.
The disk hospital is a key feature of GPFS Native RAID that

asynchronously diagnoses errors and faults in the storage
subsystem. GPFS Native RAID times out an individual disk I/O
operation after about ten seconds, thereby limiting the impact
from a faulty disk on a client I/O operation. The suspect disk
is immediately admitted into the disk hospital where it is
determined whether the error was caused by the disk itself
or by the paths to it. While the hospital diagnoses the error,
GPFS Native RAID uses the redundancy codes to reconstruct
lost or erased stripes for I/O operations that would otherwise
have used the suspect disk.
Bringing It All Together

GPFS Storage Server is a storage appliance that follows in the
GPFS tradition of being adaptable. All of the GNR software
exists separate from the existing GPFS functionality. This
means that in addition to the new RAID capabilities of GNR
you can continue to use all of the GPFS features including the
ability to mix any storage into the cluster. You can use NSD’s
on a GSS in an existing cluster and even in an existing file
system.
Already proven in the field, GPFS Native RAID (GNR) has

demonstrated the ability to succeed under very demanding
I/O workloads. GSS is a key next step in a series of next
generation software-based storage technologies.
Chapter 5
Software Defined Storage

in the Real World
In This Chapter
▶ Reviewing high-performance computing in higher education
▶ Discovering storage solutions for energy and telecommunications
▶ Building better systems with engineering analysis
▶ Examining solutions in the life sciences industry
▶ Taking a look at media and entertainment uses
▶ Studying research and development applications
I BM’s General Parallel File System (GPFS) and server and

storage systems are ideal for many different workloads
in various industries. This chapter takes a peek into several
organizations around the world that are using GPFS to solve
real-world challenges.
Higher Education
Higher education institutions must support research in
many fields over years of study. This research requires
high-performance computing (HPC) systems and storage
arrays. The United Kingdom’s Durham University Institute
for Computational Cosmology is one example of a research
institute with extreme computing and storage needs.
Established in 2002, Durham University’s Institute for

Computational Cosmology (ICC) has become a leading inter-
national center for research into the origin and evolution of
the universe, using HPC clusters to simulate cosmological
events and answer some of the most fundamental questions
in science: What were the first objects in the universe? How

do galaxies form? What is the nature of dark matter and dark
energy? Where does the large-scale structure of the universe
come from? What is the fate of the universe?
Following the success of the original DiRAC (Distributed

Research utilizing Advanced Computing) initiative, the UK
government allocated a second round of funding for a new
generation of HPC investment, known as DiRAC 2. The ICC
made a successful bid to create a new cluster, which would be
known as COSMA5.
“The strategy behind DiRAC 2 is that the clusters are designed

for different types of research problems,” says Dr. Lydia Heck,
Senior Computer Manager at the ICC. “COSMA5 is optimized
for ‘big data’ projects, while, for example, the Leicester cluster
and the Cambridge HPCS cluster are designed for optimal
high-performance inter-process communication, and the
Cambridge Cosmos system for shared memory. Running the
right code on the right platforms makes a huge difference to
efficiency, as well as enabling the UK’s universities to tackle a
wider range of projects. To make sure COSMA5 would deliver
the petascale storage and data-processing power required for
our role in DiRAC 2, we needed to put the best possible infra-
structure in place. We performed a full EU procurement exer-
cise to select the right hardware and implementation partners
to help us build the cluster, and a joint proposal from IBM and
OCF achieved the highest score of all the bids we received.
Their success was not just based on price and technical capa-
bility, but also on support and service levels — and they had
good references to support their proposal, which made us
confident that they could deliver what we needed.”
“COSMA5 is four times as fast as the previous-generation

COSMA4 cluster — achieving sustained performance of
126 TeraFLOPS at 91 percent efficiency on the LINPACK
benchmark,” says Dr. Heck. “It is also 28 times as fast as
the old air-cooled COSMA3 cluster, which we have now
retired and removed from the machine room. “Replacing
COSMA3 with a more powerful water-cooled cluster means
that our total HPC landscape is much more energy efficient:
The machine itself not only uses considerably less electricity
per FLOP, it also requires no air conditioning.”
Chapter 5: Software Defined Storage in the Real World 47
The ICC now needs no air chillers; all of the cooling require-
ments are met by the iDataPlex water cooling system. As a
result, the machine room’s power utilization effectiveness
(PUE) score has been reduced to around 1.2 — which means
that more than 80 percent of the electricity used by the room
goes directly to the clusters themselves, rather than powering
ancillary systems.
COSMA5 has also seen the ICC adopt new software for cluster
management: IBM Platform HPC, which includes IBM Platform
MPI for communication between processes and IBM Platform
LSF for workload management.
“The IBM Platform Computing software is more stable and

user friendly than our old open source software,” comments
Dr. Heck. “One of our objectives is to sell HPC capacity to
business customers, and we feel that these tools will put us in
a better position to deliver professional levels of service.”
Meanwhile, the scientific community is already working with

COSMA5. A cosmological research project known as EAGLE is
using the cluster to model galaxy formation. The UK magneto-
hydrodynamics community is also harnessing the cluster to
investigate star formation and interstellar clouds.
Dr. Heck concludes: “COSMA5 is playing an active role in the

advancement of cosmology and astronomy, helping the UK
maintain its position as a leader in these areas. We’re excited
by its potential to support academic and commercial science
projects across the country and around the world.”
Energy
The energy industry has massive processing and storage
challenges. For example, the Oil and Gas (O&G) companies
explore vast, remote areas around the world in search of new
oil reserves. Seismic imaging and ultrasound data enables
these companies to map what is beneath the surface — on
land and sea.
Sui Southern Gas Company (SSGC) Limited implemented IBM

business analytics and data warehousing solutions to help
tackle its data processing and management challenges. SSGC
is Pakistan’s leading natural gas provider, with a transmission
pipeline network more than 3,220 km long and a distribution
network that spans more than 1,200 towns. The company has
more than 2.2 million industrial, commercial, and domestic
customers and supplies approximately 390 million cubic
feet of natural gas each year. SSGC also owns and operates
Pakistan’s only gas meter manufacturing facility, with annual
production capacity of more than 750,000 meters.
To manage its transmission and distribution networks, SSGC had

more than 5,000 paper maps and was steadily adding new maps
as the scope of its operations grew. The day-to-day handling,
storage, and maintenance of these maps were a significant drain
on resources. More important, the information they contained
was outdated almost as soon as they were created, and there
was no consistent scale or format for the maps. Reconciling
operational data with mapping data to get a clear picture of
supply-and-demand issues was practically impossible: Not
only were the maps inconsistent and difficult to use, but also
there was no single source of reliable data.
SSGC wanted to be able to visualize data by geographic

location, to improve its efficiency in matching supply and
demand, and to help plan its ongoing corporate expansion.
The company also wanted better visibility of gas leakages
and unaccounted-for gas (UFG) so that it could close the gap
between the value of gas delivered to the network and the
value of the revenues generated from its sale.
SSGC implemented a business analytics warehouse solu-

tion based on the IBM Smart Analytics System 7700 (ISAS
7700). The first stage of the project was the creation of a
consolidated data warehouse. In the second stage, Infosphere
DataStage was implemented to enable seamless integration
with Cognos Business Intelligence Reports and Dashboards
for routine analysis and reporting.
The IBM solution aggregates geographic data from more than

5,000 maps with data on assets (pipelines, fittings, depots),
road and rail networks, and data from SSGC’s ERP and billing
systems. By combining real-time operational and current
geographic information, the solution enables timely, sophisti-
cated analytics that reveal previously hidden insights. Uniting
operational data with geographic data brings reporting and
analysis to life, enabling SSGC to visualize supply and demand
issues as never before. What-if analyses enable the business to
test new ideas faster and understand the results with greater
clarity, supporting key business transformation initiatives.
The IBM Smart Analytics System 7700 solution is a pre-integrated
and optimized stack of data warehouse management software,
analytics tools, storage, and IBM Power Systems servers.
Designed to deliver optimal performance and flexibility for
business analytics, the IBM solution was also fast and easy
to deploy. As a complete business-ready analytics solution, the
IBM Smart Analytics System 7700 offers simple deployment
and operation, yet provides a rich and complete stack of
technologies with the resiliency required for SSGC to analyze
data with confidence and focus on business issues rather than
on platform integration.
The key components of the data warehouse are IBM

InfoSphere Warehouse, IBM InfoSphere Business Glossary, and
IBM InfoSphere DataStage. The IBM Smart Analytics System
7700 deployed by SSGC is a pre-integrated stack of data
warehouse management software and analytics tools, with
an IBM System Storage DS5300 array and an IBM Power
740 server with IBM POWER7 processors. Tuned to deliver
optimal performance and flexibility for business analytics, the
IBM solution was also fast and easy to deploy, enabling SSGC
to focus on business issues rather than on integration.
The combined GIS and analytics solution provides a web

portal that delivers detailed, multi-layered digital maps
overlaid with business and customer information for
improved decision-making. Users can generate up-to-date
versions of pre-built reports tailored to meet the needs of
different job functions and management levels. Executive
dashboards provide at-a-glance views of performance, and
managers can use the intuitive drag-and-drop interface of IBM
Cognos Report Studio to create their own tailored reports.
The IBM solution gives SSGC faster, more accurate, and more
comprehensive reporting and analytical capabilities, helping
the company to identify hot spots for leaks and pilferage,
to visualize supply-and-demand issues, and to undertake
what-if analyses to plan more efficient processes. The creation
of a data warehouse to store operational data for analysis
and reporting has given SSGC a single version of the truth,
improving the consistency and reliability of reporting. Thanks
to the data warehouse, SSGC can also now easily leverage
information from a wide variety of sources, providing a richer
view of operational performance across multiple different
dimensions and a better understanding of customer behavior.
In SSGC’s old environment, resources were assigned manually

to workloads — all that happens automatically in their new IBM
software defined environment. User-driven reporting with real-
time data updates has replaced the laborious, time-consuming
creation of static reports. Different views and dashboards are
provided to each different type of information consumer within
SSGC. With a single trusted version of information in a centrally
managed environment and the ability to geo-reference data,
SSGC now has a scalable integration platform to handle the
variety of data, and dramatically improve the speed and clarity
of its insight into supply and demand issues.
Engineering Analysis
Engineering analysis enables design engineers to analyze
the individual components of a complete system, applying
scientific analytic principles and processes, to understand its
operation.
Infiniti Red Bull Racing uses a HPC cluster, featuring IBM GPFS,
to power its high-performance computing infrastructure for
both design applications and near real-time race analytics,
giving the racing team the edge it needs to design and run the
best cars on the track.
While races may be won by drivers at the wheel, building a

championship-winning Formula One car requires the ongoing
efforts of designers, scientists, and engineers, backed by
crucial support from a vast array of partners.
Al Peasland, Head of Technical Partnership at Infiniti Red Bull

Racing, elaborates, “Formula One regulations are becoming
more stringent every year, with governing bodies limiting the
amount of wind-tunnel and on-track test time we can use.
We rely heavily on virtual analysis and simulations, such as
computational fluid dynamics (CFD), to form the backbone of
our testing and development work.”
With CFD, Infiniti Red Bull Racing can perform virtual wind-
tunnel testing on new car designs as a first step to determining
the impact of design changes on a vehicle’s aerodynamics.
Simulation is a critical factor in analyzing design improvements
and requires huge amounts of processing power.
To support top performance for design and analytics applica-
tions, Infiniti Red Bull Racing uses an HPC cluster, featuring
IBM GPFS, which enables high-speed file access to applications
running on multiple nodes of the cluster.
The Infiniti Red Bull Racing team knew that as demand for the
data generated by the HPC cluster took off, managing the work-
load would be essential. After initially examining open source
alternatives, the team turned to IBM Platform Computing to
make better use of its enormous compute resources using
intelligent scheduling and resource allocation.
Nathan Sykes, CFD and FEA Tools Group Team Leader at

Infiniti Red Bull Racing, remarks “With the IBM Platform LSF
family of products, we have seen a 20 to 30 percent reduction
in the time it takes to complete simulations simply by being
able to design complex, interdependent workflows and
schedule individual jobs automatically. This means that we
can continually run more and more analyses significantly
faster than before. As a result, we can redesign, model, and
build a new Formula One car in about half the time that it
used to take in the past.”
With IBM Platform Process Manager, the team can manage

and control the end-to-end process flow and submission of
jobs to Platform LSF. The solution is used in conjunction with
IBM Platform Application Center, which supports flexible,
web-based submission of simulation workloads.
This sequence of jobs forms an overall design workflow,

providing a consistent and defined path for the simulations
submitted by the designers. During the process, valuable
application licenses are managed with Platform LSF License
Scheduler, which ensures optimum utilization of expensive
software resources.
“The performance improvements that we have gained with

the IBM Platform LSF family allow us to do much more in
shorter timeframes,” comments Nathan Sykes. “We can design
and test more components, and do so very quickly, which
gives us a greater chance of making the design breakthroughs
we need to build the fastest and most aerodynamically
efficient cars on the track.”
Building on its investment in the IBM Platform Computing

family of products, Infiniti Red Bull Racing is currently working
to deploy IBM Platform Analytics. Once fully operational, the
advanced analysis and visualization tool will help turn mas-
sive amounts of workload data into valuable insight and
provide powerful reporting capabilities.
In addition to using virtual testing to optimize car design,

Infiniti Red Bull Racing takes advantage of analytical applica-
tions to power faster decisions on the racing track. For this,
IBM Platform Symphony supports the processing of in-race
sensor information from the car and track to return recom-
mendations for the team on tires, designs, and strategy.
“We have seen an increase in performance with the analytics

work we are doing, both in terms of volume and speed,” notes
Christian Horner. “Instead of it taking days or hours to run
a simulation, we are now completing them in minutes. This
means that we are able to get results to the track side much
faster, which allows the team to make intelligent adjustments
and decisions during the course of a race weekend.”
Infiniti Red Bull’s software defined environment optimizes

their compute, storage, and network resources so they can
adapt to the type of work required — from structural analysis
and fluid dynamics in the race car design process — to real-
time analytics driving tactical decisions during a race.
The partnership between Infiniti Red Bull Racing and IBM

Platform Computing has helped to fuel impressive results for
the racing team and will serve as a solid foundation for future
success.
Christian Horner concludes, “IBM Platform Computing has

played a huge role in speeding up design, simulation, and
analysis, and now forms an integral part of how Infiniti Red
Bull Racing develops championship-winning Formula One
cars. For an energy drink company like Red Bull to take on the
might of some huge engineering companies and beat them is
something that we are all fiercely proud of. We rely on part-
nerships such as the one with IBM Platform Computing to
power engineering and design excellence, and to sustain the
kind of success that we have enjoyed in recent seasons.”
Life Sciences
The life sciences industry encompasses numerous fields
including agriculture, biochemistry, food science, genetics,
health and medicine, medical devices and imaging, and
pharmaceuticals, among others.
The University of Bonn uses IBM technologies, including

clustered IBM BladeCenter servers, IBM System Storage
arrays, and IBM GPFS, to support critical research in the
field of genetics.
The University of Bonn’s Institute of Medical Biometry,

Informatics and Epidemiology is a world-leading medical
research center in the field of medical genetics research. The
Institute makes significant use of computational models and
carries out complex statistical and analytical calculations
on large data sets. To help gain new insights into important
research topics, the institute had set up a small cluster
solution with 14 nodes. As the number and scope of projects
and the amount of data grew by orders of magnitude, this
solution became increasingly outdated. The Institute looked
for a cost-efficient yet powerful computational platform that
would enable more complex operations at higher speed.
With limited space available in its server room, the Institute

required a high-performance computing system with a small
physical footprint. The second key requirement was for
high-speed connectivity between the computational cluster
and the data storage environment. Finally, the Institute
needed a highly scalable file system that could handle the
enormous — and growing — amounts of data that genetics
researchers expect to be able to manipulate.
“The only file system we knew on the market that would be

capable of scaling as required was IBM GPFS (General Parallel
File System),” says Waldemar Spitz, System Administrator
at the Institute of Medical Biometry, Informatics, and
Epidemiology.
The Institute deployed a total of 26 IBM BladeCenter HS22 serv-

ers and 8 IBM BladeCenter LS42 servers, providing 34 cluster
nodes with 504 processor cores in total, supported by 1.8TB
of main memory. Two IBM System Storage DS3400 arrays provide

a total of 40TB of storage capacity for the IBM BladeCenter
cluster — and the Institute is actively adding more capacity.
To satisfy the institute’s networking and storage performance

needs, IBM recommended IBM System Storage DS3400 devices,
connected via Fibre Channel to IBM System x3650 M2 servers,
set up as IBM General Parallel File System I/O servers, which
are linked via InfiniBand connections to the high-performance
computing cluster.
Today, researchers in two teams use the computing resources

provided by the IBM solution and the Institute also makes the
resources available to guest researchers from around the world.
New research enabled by the IBM cluster covers all aspects of
genetically complex diseases and population genetics.
The researchers not only benefit from the massively increased

performance, but also experience fewer interruptions. “The new
IBM cluster solution runs in a more stable manner, especially
under high load,” says Spitz.
The cluster solution from IBM has also made management

much easier for system administrators. In the past, the team
used customized scripts to manage its cluster. Today, the
Extreme Cloud Administration Toolkit provides standardized
tools and methodologies for cluster administration, increasing
the systems management efficiency at the Institute. This is a par-
ticularly important benefit, because it frees up researchers from
routine administration and enables them to focus on science.
The IBM BladeCenter cluster processes significant volumes

of useful medical research data, evidenced by the fact that
the Institute is now expanding its storage environment with
an IBM System Storage DCS3700 array and an IBM System
Storage DCS3700 Expansion Unit to 400TB of net capacity.
Designed for applications with high-performance streaming
data requirements, the IBM DCS3700 offers optimal space
utilization, low power consumption and high performance.
With up to 60 SAS drives in just 4U of rack space, it can reduce
operational costs for capacity-intensive applications. And with
up to 4000 MBps in sustained drive reads, the IBM DCS3700
storage system is equally adept at delivering throughput to
bandwidth-intensive applications.
The use of GPFS will make it easier for the Institute to
continue expanding the storage environment, as the file
system has practically unlimited scalability.
Media and Entertainment

Media and Entertainment (M&E) companies create and store
tremendous amounts of data through various stages that
include capture, post-production, editing, and archiving.
Speicher M1 GmbH, based in Bremen, Germany, offers

cloud-based media asset management services. The company
provides a collaboration platform to host and manage video
data for customers in the movie and video industry. Secure
archiving on durable media is crucial to protect media
assets and enable long-term monetization of movie footage.
The industry faces enormous challenges around archiving
and indexing large volumes of digital video data over long
periods, prompting many companies to outsource data
hosting.
Peter Flory, Managing Director of Speicher M1 GmbH,

comments, “Our customers typically work in large, distributed
teams. Providing fast access to video data and supporting
effective collaboration among team members are key
priorities, yet most managed cloud solutions offer data
storage only.”
Speicher M1 saw a gap in the market for an end-to-end

platform combining digital media storage, management and
distribution. There was an opportunity for the company to
distinguish itself from competitors by offering a cloud-based
service for full-lifecycle management of media assets. Speicher
M1 recognized that capturing as many new customers as
possible would also depend on competitive pricing. “Our
research showed that many of the top service providers were
charging in excess of one hundred euros a month per terabyte
of data,” notes Flory. “We aimed to offer our full solution at a
much more competitive price.” Speicher M1 decided that IBM
offered the optimal platform for rapid, secure, and cost-effective
management and deployment of cloud-based applications and
services.
The cloud platform supporting the new solution is based

on four IBM System x 3650 class servers and an IBM System
Storage DS3512 Express with an IBM System Storage EXP3512
Express Storage Expansion Unit.
Speicher M1 implemented an IBM System Storage TS3500

Tape Library with IBM TS1140 Tape Drives to provide long-
term archiving capabilities. The company uses IBM Tivoli
Storage Manager to provide hierarchical storage management
for its tape storage, fully automating video archiving on tape
drives. The IBM servers and storage systems are connected
using IBM System Storage SAN24B-4 Express SAN switches
featuring fast 8 Gbps Fibre Channel links. For improved perfor-
mance and high availability, Speicher M1 operates its systems
in a cluster configuration, running Red Hat Enterprise Linux
and using IBM GPFS.
The cloud-based platform provides a host of industry-

specific features which support efficient processing and
collaboration among distributed teams. These include
advanced search and filter options that enable customers
to easily navigate through extensive footage. Customers can
access the service via an online portal, making it quick and
easy for teams to manage, edit and market projects regardless
of location.
Speicher M1 has gained a cost-efficient archiving and media

asset management platform, which will help it to better sup-
port customers in the movie and video industry. Combining
different storage technologies allows the company to offer
its customers an integrated, high-performance solution, at a
significantly lower rate than competitors.
Speicher M1 now has a fast and secure end-to-end media

asset management platform running on a private cloud, which
enables a full-service solution at a fraction of the price of
competing storage-only solutions. Powerful software defined,
policy-driven automation dynamically moves large media
asset data through tiered storage pools, based on perfor-
mance or cost requirements. Data migrated to tape remains
visible in the file system and is directly accessible to end
users while lowering storage costs by 80 percent and giving
Speicher M1 a significant competitive advantage in their
service offerings.
“By using a mix of IBM disk and tape storage, we can balance
price and performance, improving the cost-efficiency of the
solution significantly,” says Flory. “This approach allows us
to offer a high-performance solution at a competitive price,
which we believe puts us ahead of our competitors.”
The new solution provides support throughout the entire

media asset lifecycle, and features easy integration with web
shops and other third-party systems. This enables media
companies to process, distribute and market their videos with
minimal effort.
“Our customers can preview footage online without downloading

large video files,” says Flory. “They can categorize and discuss
clips while working on the movie. Finally, they can share the
finished movie with partners, host online screenings or even
connect their archive directly to web shops. This provides com-
panies with one central platform for all video production and
distribution needs — without large upfront investment costs.”
Research and Development

Research and development (R&D) activities in practically
every industry generate large amount of data and have
intensive computational requirements.
Jülich Supercomputing Centre (JSC) is part of Forschungszentrum

Jülich GmbH, the largest research center in Germany. JSC
supports research into a wide range of fields including funda-
mental physics, life sciences, climate change, and energy by
providing the powerful infrastructure and technical expertise
needed to run large and complex simulations quickly and
effectively.
JSC had to put major effort into managing its previous storage
infrastructure, which relied on expensive “hardware defined”
data protection in the form of dedicated RAID controllers.
Replacing that with a software defined storage environment
(IBM GPFS Storage Server) allowed them to utilize the abundant
computing power available across the GPFS cluster to be
redirected on demand, to dynamically support RAID rebuilds
at much higher speeds and at half the cost, compared to
conventional hardware defined storage systems.
To remain at the forefront of global research, JSC deployed an

IBM Blue Gene/Q supercomputer with 28 racks, at the time the
fastest and most energy-efficient supercomputer in Europe.
The center uses the IBM supercomputer for research into
complex systems such as the human brain.
Modeling complex systems requires huge amounts of

data — especially the so-called “scratch data” that is used
during simulations, but not stored for the longer term. JSC
must provide the IBM supercomputer with reliable, rapid
access to scratch data.
JSC also needed a new storage infrastructure to match the

capabilities of the IBM supercomputer. The most important
requirements for this new storage infrastructure were reliability
together with high bandwidth and capacity.
JSC decided to implement an IBM System x GPFS Storage

Server (GSS) solution, eliminating the need for separate physi-
cal storage controllers. GPFS uses GPFS Native RAID software
to deliver not only outstanding throughput, but also extreme
data integrity, faster rebuild times, and enhanced data
protection.
Chapter 6
Ten Ways to Use

Software Defined
Storage
In This Chapter
▶ Looking at innovative uses for software defined storage in your
organization
T here are many different computing and data challenges,

each with a unique set of requirements related to scale,
performance, retention, access, availability, and capacity.
This chapter give you ten (okay, seven) innovative uses for
software defined storage using IBM GPFS (General Parallel File
System) that can help you solve your storage challenges.
Improve Application
Performance
Data-intensive applications are defined by the fact that they
need to read or write a large amount of data to get the job
done. Speeding up data-intensive applications is easy, use a
faster storage infrastructure. In reality, it may not be so easy
because many storage solutions don’t scale efficiently. When
the storage device does not scale the only way to improve
I/O performance is to use many small data containers and
modify your applications to utilize these separate containers
concurrently. If you don’t spread the data for the project
across multiple containers adding containers can create data
hot spots. A data hot spot occurs when you have a high level
of I/O required by a set of data on a single NAS appliance.
GPFS eliminates hot spots by spreading data across all of the
available storage hardware.
GPFS is designed to support I/O workloads run by a large

number of processes and servers concurrently accessing a
common set of file data. GPFS stores and accesses file-based
data in blocks that are striped across many disks. Striping
data provides the best performance for whatever set of files
you are using because you can get the full performance of the
underlying storage.
For many storage products you have to use separate c ontainers

to provide better overall throughput. The problem with this
approach is that you partition up your data, greatly complicating
management tasks, and it is difficult to keep all of your hardware
busy. For GPFS, the opposite is true. GPFS allows you to fully
leverage the performance of all of the underlying storage
hardware. It does this by spreading the file data over all of the
available storage all the time. This means that you don’t have
idle disks and more importantly not wasting money.
For example, in the entertainment industry, any time a celeb-

rity makes the news their fans want to download pictures and
movie clips of that celebrity, causing a set of data associated
with that celebrity to become “hot.” To support these types
of bursty workloads, some storage products use techniques
such as file replication or RAM caching.
The problem is that replication-based products require extra

space for files and extra bandwidth to make the copies. Also,
in the case of streaming media, the load against each file
varies with time, and replication has to be done continuously.
And RAM caching is cost-effective only for very large servers
and thousands of parallel media streams.
Making all of this work takes advanced data integrity and

sharing methods. GPFS is very “fair,” providing equal access
to the data for all servers and processes. The efficient use of
storage space along with fault tolerance ensures data integrity
in GPFS-based solutions.
Metadata in GPFS is distributed similar to data. There are two

aspects to metadata attribute storage and data consistency. As
Chapter 6: Ten Ways to Use Software Defined Storage 61
with other data, metadata is spread across all available
storage and metadata management is distributed across
the entire cluster. Also, many metadata-intensive workloads
perform much better with GPFS, leveraging its distributed
metadata and load balancing features. Applications that
require dynamic load balancing need a file system that has
excellent I/O performance and is very reliable. GPFS performs
like a local file system, with the added advantage of flexibil-
ity, increased scalability, and the reliability of a clustered file
system.
GPFS enables all of the servers in a cluster to access all system

data equally and metadata in parallel. This improves perfor-
mance for metadata-intensive applications by speeding up
application I/O operations. Since GPFS allows any server in a
cluster to read from or write to any of the disks, applications that
perform concurrent I/O can achieve very high data access rates.
GPFS is POSIX compliant, so you don’t need to modify

applications to implement an enterprise class file-based
storage solutions layered on top of GPFS. Applications can
leverage the concurrent data access capabilities of GPFS and
achieve horizontal scaling and load balancing. Because the
file system is distributed over all the GPFS servers, you can
run the same version of a web server application, for example,
which is accessing a database on all GPFS cluster servers.
This allows an update to the database to be seen on all
servers as soon as one server inserts a record.
Many application workflows, for example, in weather forecasting

and financial analysis, operate on a large number of individual
parts that run as parallel jobs. If one component fails, it could
impact the total time taken by the job and reduce application
performance.
Using GPFS, I/O overhead can be significantly reduced —

especially those dealing with storing and retrieving the appli-
cation state data from a checkpoint — in order to restart the
failed server (or replace it with a new one) and resume the
job. With the GPFS massive single namespace feature, you
can easily add a new server to replace a failed server and
restart only the failed job from the last saved checkpoint. In
a GPFS-based storage infrastructure supporting a multi-part
application running on multiple servers, the application data
is not tied to any single server.
GPFS provides a highly scalable, low-latency, high-performance,

reliable file system for large-scale storage infrastructures, with
capabilities for distributed parallel data streaming and no single
point of failure. It’s not uncommon to have GPFS file systems of a
petabyte or more, containing hundreds of millions of files.
Data Sharing and Collaboration

A single GPFS file system can support a mix of workload types
using the same disk or even different types of storage for two
files in the same directory.
With the continuing rise of data-driven applications, the gap

between performance requirements and system capabilities
for storage, data management, and I/O access has become
much wider, and the need for distributed storage and
processing has grown. Some examples of these data-driven
applications include
✓ Applications that synthesize and process results

generated by large scale simulations
✓ Business transactions
✓ Online trading and data processing
✓ Long-running compute and analysis jobs
✓ Applications that analyze large datasets for
• Medical imagery
• Seismic data for oil and gas exploration
• Industrial design
• Internet collaboration
• CRM (customer relationship management)
• Social marketing trends
• Market analysis
In many cases, datasets are generated at multiple sites. For

example, investigation of alternative production strategies for
optimizing oil reservoir management requires large numbers
of simulations using detailed geologic descriptions. The
output from reservoir simulations is usually combined with
seismic datasets that are often stored on different systems
and in different locations, in order to better predict the
reservoir properties and drive new simulations.
In the automotive and manufacturing industries, product

design and innovation often happen across the globe among
distributed teams. CAD (computer aided design) and other file-
based data from multiple teams located in different geographic
sites may need synthesis and collaboration to integrate various
aspects of a design. Such data is often fragmented into small
storage islands that use different technologies and are managed
by independent teams, each with their own local use policies.
Cross-site projects are often hindered by this inconsistent
approach to technology and management. Also, many of the
technologies employed are not designed for enterprise scale
computing and teams struggle to expand their IT environments
to meet the growing needs of cross-site collaboration.
With an enterprise-wide GPFS environment, you can achieve

cost-effective and efficient collaboration with features such as
a common file system and massive global namespace across
computing platforms. Users can seamlessly access data from
any storage cluster server without having to first transfer the
data from another location. This streamlines the collaboration
process and is more cost effective and energy efficient because
enterprises don’t have to purchase additional disk space to
store duplicate files. By pooling storage purchases, you can
build a much larger common shared data infrastructure. In
addition, the data is available in a highly parallel manner,
making access to massive amounts of data much faster.
Information Life Cycle

Management
It’s been estimated that as much as 90 percent of all file data
is never needed again after it is initially created and, unlike
fine wine, most data gets less valuable the older it gets. With
petabytes of data and billions of files, it isn’t practical to just
ask each application group to “clean up their stuff.” Plus,
government regulations often require you to keep certain data
around for many years. Solving this problem requires automation,
and GPFS provides a set of tools designed to help you.
GPFS helps simplify storage administration through its policy-

based automation framework and Information Life Cycle
Management (ILM) feature set. GPFS policy-based ILM tools

can be used to manage sets of files and pools of storage, and
automate the management of file data. Using these tools,
GPFS can automatically determine where to physically store
user data, regardless of its placement in the logical directory
structure. Storage pools, Fileset, and user-defined policies
provide the ability to match the cost of storage resources to
the value of your data.
Storage pools allow you to create groups of disks within a file

system. You can create tiers of storage by grouping your disks
based on performance, locality, or reliability characteristics.
For example, one pool could be solid state disks (SSDs) and
another could be more economical SATA (Serial ATA) storage.
A fileset is a sub-tree of the file system namespace and

provides a way to partition the namespace into smaller,
more manageable units. Filesets provide an administrative
boundary that can be used to set quotas, take snapshots,
and be specified in a policy to control initial data placement
or data migration. Data in a single fileset can reside in one or
more storage pools. Where the file data resides and how it is
migrated is based on a set of rules in a user-defined policy.
GPFS has two types of user-defined policies
✓ File placement
✓ File management
When a file is created, GPFS needs to know where to put it.

This is done by using file placement policies that direct file
data as files are created to the appropriate storage pool. You
can create file placement policies based on anything GPFS
knows about a file when it is created, including filename and
the user who is creating the file.
After a file has been created, GPFS knows much more about
it. In addition to the attributes available when a file is created,
GPFS now knows additional information including the size of
the file, how long it’s been since someone accessed the file,
and whether or not it’s been changed. Policies that operate on
existing file are called file management policies and allow you
to move, replicate, or delete files. You can use file manage-
ment policies to move data from one pool to another without
changing the file location in the directory structure. One
popular use for file management policies doesn’t involve
moving data at all — you can use it for reporting. The policy
syntax is very powerful, allowing you to generate custom
reports, for example, on the type of files using the most space.
On Linux and AIX, you can use similar tools to get this infor-
mation, but the policy engine is very fast — it can look at the
metadata of millions of files per second.
ILM tools need to have rich features, be automated, and be

capable of operating on very large data sets to be useful when
you are storing petabytes of data. The GPFS ILM toolset is well
suited to this environment and is capable of managing billions
of files.
System Administrators, Take

Back Your Weekend
Downtime — whether planned or unplanned — is costly.
Making sure your applications are always available requires a
robust, high availability solution, so applications can continue
to run in the event hardware fails, capacity is expanded, or
hardware and software is upgraded.
The core GPFS software is fault tolerant. If a server or even a

storage system fails, the other servers can continue to access
the data. It does this by continuously monitoring the health
of the cluster and file system components. When a failure is
detected, the appropriate recovery action is taken automati-
cally. Extensive logging and recovery capabilities maintain
metadata consistency when application servers holding locks
or performing cluster services fail.
For additional protection, you can have GPFS create two or three
synchronous copies of the file system metadata and/or the file
data. When data is replicated, you can use either copy of the
data for reads in an active-active mode. You can even tell GPFS
which copy to read from, in order to keep read access local if
you are replicating across sites. This can help for read-intensive
workloads and to keep certain traffic off the WAN. When both
copies are located in the same data center, for example, reads
come from both copies, thereby doubling your read bandwidth.
Software defined storage introduces new ways to manage data

like file-based replication in an active-active configuration.
This gives you flexibility to keep data safe and be able to use
it at the same time, across various types of hardware.
Hadoop MapReduce Workloads

There is an industry trend toward taking a MapReduce
approach to analyzing large amounts of data, sometimes
referred to as big data. The expected cost benefit of using
MapReduce is realized by achieving high I/O throughput
on very inexpensive hardware. The approach is to leverage
the internal I/O performance of a server, while being able to
task a bunch of these servers to solve big problems. To run
a workload of this type requires storage software that can
support this hardware architecture, and provide the right
interface for MapReduce applications to find the right data.
Both the Hadoop Distributed File System (HDFS) and GPFS are
designed to provide a storage platform for data supporting
MapReduce workloads, on large-scale standard hardware
consisting of thousands of servers.
MapReduce is a programming and data organization model

for processing large data sets in parallel on a distributed
computational cluster.
HDFS and GPFS both provide the basic storage tools needed
for MapReduce workloads, but that is where the similari-
ties end. HDFS is a basic storage solution for Map Reduce,
whereas GPFS is an enterprise storage software solution that
supports MapReduce. Some limitations of HDFS include
✓ Centralized master-slave architecture

✓ No file locking
✓ File data stripped into uniformly sized blocks that are
distributed across cluster servers
✓ Block-level information exposed to applications
✓ Simple coherency with a write-once, read-many model
that restricts what users can do with data
GPFS features include
✓ High-performance, shared-disk cluster architecture with

POSIX semantics
✓ Distributed metadata, space allocation, and lock
management
✓ File data blocks striped across multiple servers and disks
✓ Block-level information not exposed to applications
✓ Ability to open, read, and append to any section of a file
GPFS includes a set of features that support MapReduce work-

loads called GPFS File Placement Optimizer (FPO). GPFS-FPO is
a distributed computing architecture where each server is self-
sufficient and utilizes local storage. Compute tasks are divided
between these independent systems and no single server waits
on another. GPFS-FPO provides higher availability through
advanced clustering technologies, dynamic file system
management, and advanced data replication techniques.
In addition, GPFS supports a whole range of enterprise data

storage features, such as snapshots, backup, archiving, tiered
storage, data caching, WAN data replication, and management
policies. GPFS can be used by a wide range of applications
running Hadoop MapReduce workloads and accessing other
unstructured file data.
In laboratory tests, benchmarks demonstrate that a GPFS-

FPO-based (modifies system) system scales linearly so that a
file system with 40 servers would have a 12GB/s throughput, and
a system with 400 servers could achieve 120GB/s throughput.
Cloud Storage
Cloud storage provides a scalable, virtualized infrastructure as
a service, hiding the complexity of fine-grained resource man-
agement from the end-user. According to IDC (International
Data Corporation), the amount of information in the world
is set to grow 44-fold in the next decade, with much of that
increase coming from the rise in cloud computing.
Software defined storage systems running GPFS are well

suited to address cloud storage requirements given their
extreme scalability, reliability and cost efficiency. Software
defined storage systems can scale to thousands of servers
while supporting hundreds of GB/sec of sequential through-
put and can be configured using lower-cost standard parts
without the need for costly enterprise-class storage solutions,
such as storage area networks (SANs).
Unlike many cloud storage solutions available today, GPFS

supports traditional applications that rely on POSIX file APIs
and provide a rich set of management tools. A cloud storage
stack built on a POSIX-based cluster file system makes it
easy to support existing applications by providing a scalable
infrastructure that doesn’t require application modifications.
Cloud storage is shared across different classes of applications

so standard file semantics are important. The standard interface
needs to support new workloads including MapReduce style
applications so you don’t have to use separate point solutions
for these workloads. For mixed workload environments which
require access to large amounts of data in a cloud environment,
GPFS can help leverage standard components, along with its
POSIX-complaint interfaces and support for the latest tech-
nologies, such as InfiniBand (IB) Remote Memory Data Access
(RMDA).
Enterprise Analytics
For enterprise analytics, like detecting credit card fraud or
generating customer specific marketing offers, scalability,
reliability, and ready access to data is critical. Analytics is a
data-driven application that delivers measureable business
value.
The IBM Smart Analytics System is a pre-integrated analytics

system designed to deploy quickly and deliver fast time-to-value.
Engineered for the rapid deployment of a business-ready
solution, the IBM Smart Analytics System includes the following
features
✓ Powerful data warehouse foundation

✓ Extensive analytic capabilities
✓ Fully integrated, scalable environment
GPFS is a core component of IBM’s Smart Analytics offering

providing a high availability data storage platform. The IBM
Smart Analytics System core warehouse servers have an
active-passive high-availability configuration whereby data is
available even in the case of server failures. The system can
be scaled up as data grows to meet new business needs such
as intelligence, smart analytics, and data warehousing.

TSW03232ESES-SWGdefinedStorage IBM PDF

Uploaded by

Copyright:

Available Formats

TSW03232ESES-SWGdefinedStorage IBM PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

TSW03232ESES-SWGdefinedStorage IBM PDF

Uploaded by

Copyright:

Available Formats

Compliments of puting Edition

IBM Platform Com

IBM Platform Computing Edition

by Lawrence C. Miller, CISSP

LIMIT OF LIABILITY/DISCLAIMER OF WARRANTY: THE PUBLISHER AND THE AUTHOR MAKE

Chapter 2: What Is Software Defined Storage?. . . . . . . 17

Chapter 3: Digging Deeper into IBM GPFS. . . . . . . . . . . 23

Chapter 4: Getting to Know the IBM GPFS

Chapter 5: Software Defined Storage in the

Chapter 6: Ten Ways to Use Software Defined

Project Editor: Carrie A. Johnson

IBM General Parallel File System (GPFS) provides online

About This Book

Icons Used in This Book

This icon points out information that may well be worth

You won’t find a map of the human genome or the blueprints

E nterprise users and data consumers are churning out vast

This chapter provides some background on digital storage

Data Access and Management

Structured data refers to data that’s organized, for example, in

To address this problem, many IT managers today are adding

Many data centers have become victims of “filer-sprawl” — IT

✓ Rising administrative costs to manage file-based data

Data access and management is critical to an efficient com-

Three Important Functions

✓ Store data (intermediate and final)

Storage administrators have always recognized the need for

However, storage administrators are now increasingly focused

✓ CPU speed performance has increased 8 to 10 times.

One result of this storage bottleneck is that your applications

Defining Types of Storage

✓ Fibre Channel (or Fibre Channel Protocol, FCP)

Fibre Channel and iSCSI interfaces, as well as SAS, are

Block-based storage systems are commonly implemented as

DAS can provide basic data protection functions (for example,

Storage area networks

Server 1 Server 2 Server 3

Figure 1-3: A SAN connects a storage array to servers via a dedicated

A LUN (Logical Unit Number) is an identifier assigned to a

The IBM XIV Storage System is an example of block-based

SANs are commonly used in mission-critical or high-transaction

✓ Fast speeds: SAN speeds are increasing dramatically due to

✓ Management: Processing and storage are managed

Compared to other storage systems, SANs can be relatively

✓ SMB (Server Message Block) or CIFS (Common Internet

NAS appliances are relatively easy to deploy, and client access

Filer 1 Filer 2 Filer 3 Filer 4 Filer 5 Filer 6 Filer 7 Filer 8

Storage Storage Storage Storage Storage Storage Storage Storage

Figure 1-4: F ilers often lead to “islands of storage” in the datacenter.

File System Object

Files SPREAD GAMES File, image, video Data

A container stores the actual data (for example, an image

Object-based storage is commonly used for cloud services

Time to explain RAID

✓ RAID 4 (block-level striping with in use today, but with growing

Hard Disk and SSD Technologies

SATA drives are typically used in desktop and laptop com­

200 and 400GB

SAS drives are commonly used in servers (DAS) or SANs. SAS

Figure 1-3: A SAN connects a storage array to servers via a dedicated

Figure 1-4: F ilers often lead to “islands of storage” in the datacenter.

SATA drives are typically used in desktop and laptop com

Figure 2-1: I BM GPFS delivers extreme I/O performance.