0% found this document useful (0 votes)
110 views

Dell Networking

Uploaded by

Freddy Vergara
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
110 views

Dell Networking

Uploaded by

Freddy Vergara
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 27

Disclaimer

This presentation contains references to certain features, functionality, enhancements,


or other technology that may not be currently available. This information is intended to
outline our general product direction and should not be relied upon in making current
purchasing decisions. These references are: i) for information purposes only, ii) may not
be incorporated into any contract, and iii) do not constitute a commitment, promise or legal
obligation to deliver any material, code, or functionality. The development, release and
timing of any features, functionality, enhancements, or other technology described remains
at the sole discretion of Dell Technologies.

Last update of this deck: November 2023

Internal Use - Confidential 1 © Copyright 2022 Dell Inc.


The Fabric of AI:
SONiC Fueling
Network-Centric Computing
Director of System Engineering, Networking
Sandeep Madhavan
November, 2023
AGENDA Generative AI and Ethernet Fabric

Vision and Strategy

 Enterprise SONiC distribution by Dell Technologies

 Dell Networking for Generative AI

Road Ahead

Internal Use - Confidential 3 © Copyright 2022 Dell Inc.


Ethernet Proliferation
Powering largest AI Cloud fabric deployments

~600M
Ethernet ports shipped annually!

Reference: Presentation by Ram V, Broadcom at OCP Global Summit

4 © Copyright 2022 Dell Inc.


What makes AI
Networking
unique?

• GPU to GPU
Communication Drives
higher bandwidth flows

• Bursty traffic

• Links are saturated in


Micro-seconds

• Training jobs run for long


periods

• Tail latency impacts job


completion time
OCP Keynote by Alexis Bjorlin at 2022 Global Summit

5 © Copyright 2022 Dell Inc.


GenAI Infrastructure Building blocks – Compute Backend (GPU)
Fabric
• Objective: GPU to GPU connectivity to execute an
AI/ML training or inference job. This fabric is where
GPUs are going to perform hyper-parameter
optimizations

• Fabric Highlights
– Dedicated fabric for GPU <-> GPU communication.
– Model training and inferencing traffic
– Ethernet solutions evolving as a preferred choice
– Performance approaching InfiniBand specs
– Each GPU-Server will have 8x400G or 8x(2x200G)
connectivity to leaf switches.
– NIC is connected to GPU & CPU
– Software Requirements :
▪ Low latency fabric
▪ High Radix switches
▪ Lower tail latency

6 © Copyright 2022 Dell Inc.


GenAI Infrastructure Building blocks – Frontend Fabric
Storage Fabrics

• Objective: Storage fabric provides access to large-scale shared storage


infrastructure. This storage is used as a shared resources for GPUs to
communicate hyper-parameters during AI/ML training or inference jobs

• Fabric Highlights
– Fabric for GPU to storage server communication.
– Typically, 25G/100G connectivity with ethernet solutions

In-band/Access/Cluster Fabrics

• Objective: This fabric is used to distribute the AI/ML jobs on to the Data
Center back-end network on GPU. In-band Management prioritizes,
batches, and provides/allocates the necessary resources (GPUs, Storage,
Network) for AI/ML applications.

• Fabric Highlights
– Fabric for managing the AI/ML jobs assignment on GPUs
– Typically, 25G/100G connectivity with ethernet solutions
– Multitenancy use-cases

7 Copyright © Dell Inc. All Rights Reserved.


GenAI Infrastructure Building blocks – OOB Use case
• Objective: OOB Fabric provides management for
GPU/Storage servers, Ethernet / InfiniBand
switches, appliances (firewall, load balancer) etc.

• OOB network on server use iDRAC interface to


read temperature/thermals, CPU/GPU utilization,
miscellaneous sensor information.

• 1G ethernet connectivity solution with basic L2/L3


features

8 © Copyright 2022 Dell Inc.


Delivering Ethernet Solutions across all use cases within AI
Fabrics
Bringing it
ALL
together –
AI fabric
Back-End (GPU Fabric)
has most demanding
requirements for raw
performance, lossless
attributes and lowest
latency
Front-End fabrics
support application
traffic, storage access
and connection to the
general network
OOB Mgmt Network for
administration and fabric
management

9 © Copyright 2022 Dell Inc.


Ethernet evolving to be the preferred choice for
backend AI fabrics
• Market inflection points for Ethernets powered by AI fabrics
– Availability of High Radix switching with next-Gen silicon technologies – 64x400G (25.6T), 64x800G(51.2T), 102.4T…

– Improved congestion monitoring, flow control, and Transport (RoCEv2) protocol availability in NOS

– Community effort to drive Ethernet Standards – Ultra Ethernet Consortium

– Desire for no-vendor lock-in infrastructures – InfiniBand (Nvidia), Ethernet (Commodity vendor market)

– Silicon and supplier diversity

– Lower Total Cost of Ownership (~3x lower)

– Latency improvements with next Gen Silicon from 800ns to 200ns

10 © Copyright 2022 Dell Inc.


Networking Evolution and Market Trends
KEY MARKET TRENDS TA R G E T I N D U S T R Y
• Service Provides (T2 Cloud, Infra, Software)
• Large Enterprises
• Higher Ed & Research • Automotive
• Retail, eCommerce • Manufacturing
Data boom Distributed infra Next gen • Semiconductors
5G & IoT AI/ML Zero Trust
at Edge Cloud & On-prem automation • Telco

CUSTOMER NEEDS- Control, Extensibility and Agility

• Cloud economics • Multi vendor • Open ecosystem – Software & Tools • 25G in Rack (50G,100G for select workloads)
• High operational agility • Deep analytics and telemetry • World-class Support • Fabric transitions to 400G
• Application Centricity • Green Initiatives/sustainability • Standards for Interoperability • Multi-cloud strategy
• Scalability & Security • Investment Protection • Ease of network management • Network disaggregation

DELL TECHNOLOGIES PORTFOLIO

Next-Gen Data Center SmartNIC/DPU for co-lo Virtual Edge platform/uCPE Integrations & Ecosystem
& Edge switching and faster data processing at Edge for Edge computing Partner Solutions

Dell Enterprise SONiC < -- > Dell SmartFabricOS10 with SmartFabric Services

11 © Copyright 2022 Dell Inc.


Current Solutions in the Market
Locked in and complicated
• Control
– Focused on locked-in solution

• Choice
– Not much control over the choice of software for use cases

• Cost
– Expensive
– Locked a single source supply and short lead times

• Convenience
– Automation is an afterthought and dependency on the proprietary controller. Almost never works for
customers but they throw people at the problem.

• Customization
– Architecture not viable for quick customization to achieve an unique outcome
12 © Copyright 2022 Dell Inc.
Networking Following the
Footsteps of Compute Evolution
Linux is Born Linux Distros Red Hat offers Enterprise Linux Red Hat leverages Linux
Server OS Evolution

Capitalizes on market demand Spans eco system solution

1991 1995 2002 2015

Slackware, SuSe, Enterprise ready, reliable Ansible, OpenShift,


Debian, Red Hat Linux, etc. and supported solution OpenStack

OPEN SOURCE OS OPEN SOURCE OS (COMMERCIAL)

Dell Technologies offers Eco system around Expanded


Open Networking (ON) Open Source NOS
Enterprise SONiC Enterprise SONiC Use cases
NOS Evolution

2014 2016 2020 2021 2022

Dell Technologies onboards Microsoft contributes SONiC to OCP, An Enterprise ready and Ansible, Prometheus, Edge: Retail/Branch office
3rd party NOSs (e.g., Big Switch, Dell Technologies continued partnership supported distribution of SONiC Telegraf, Augtera, Apstra DCI BGP EVPN Multi-site
Pluribus) on Dell platforms with OCP to develop ONIE, SAI and SONiC

OPEN NETWORKING OPEN SOURCE NOS OPEN SOURCE NOS (COMMERCIAL)

13 © Copyright 2022 Dell Inc.


SONiC – Innovation of the industry > 1 Vendor Solution

Dell Technologies Participation in


upcoming SONiC focused Industry
Conferences

Conference Date Location


SONiC Workshop September 6, India
India 2023 2023

OSS Europe, September Spain


SONiC Mini 18, 2023
Summit

Open Compute October 17- San Jose,


Project, , SONiC 19, 2023 CA
Mini Summit

Open Networking October 24- New York


User Group 25, 2023 City
(ONUG)

14 © Copyright 2022 Dell Inc.


15 © Copyright 2022 Dell Inc.
Dell Enterprise SONiC

Vision & Strategy


VI SI ON STRATE GY
• Offer commercial distribution of SONiC
• Cloud-inspired NOS to provide Enterprises
with reliability, flexibility and disaggregation • Enable one unified NOS across the
product portfolio
• Be the Linux of Networking
• Enterprise enablement – Extending SONiC from
• Flexible standards based Open interfaces DC to Edge, Telco & Enterprise DC features
• Value proposition :
• E2E Support –
SW, HW, Ecosystem technologies • Ecosystem partner technologies
• Global supply chain • Documentation and Training
• Predictable Roadmap

Customers want full control and choice of the technology stack – HW, SW, and ecosystem tools, at optimal cost while avoiding
vendor lock-in

Single NOS in a multi-vendor environment


Ecosystem & Apps Invest in a future-proof network with confidence
Choice of Support

without reliance on a single vendor roadmap

Open Interfaces Build a fabric to fit your needs


Broader portfolio options enable greater flexibility
in fabric use cases and architectures
NOS Standardization Choice, flexibility and simplicity with
Enterprise support
Unified software allows simplicity and ease of
use
+ Dell qualified OpenHW platforms 16 © Copyright 2022 Dell Inc.
Dell Customer Communication - Confidential

Enterprise SONiC Distribution by Dell Technologies


• Enterprise grade features and hardening
Enterprise Ready
• Feature visibility

Enterprise Support Services • Global 24x7 ProSupport and ProDeploy


and cloud Partner
datacenter ecosystem
• 1M+ lines of code | 5K+ defect/bug fixes
features Community
• Ansible modules | User groups
Contribution
• User documentation
• Cloud use case
Enterprise Use case driven • Enterprise use case
• Edge use case
Tested & SONiC
Automation • Ecosystem partners: Augtera, Apstra, Dorado
validated by Dell Integrations • Open-source solutions: Ansible Collection certified
Technologies by RedHat, Telegraf, Prometheus, OpenStack

Automation • Open source, open interfaces • Automation ready


and Visibility • Telemetry and Deep analytics • Container-based architecture

Global Tools and


PowerSwitch • 3rd-party container management
HW/SW Utilities
HW
support
• Virtual demos, Hands-on Labs, Fabric Design Center,
Education Technical papers, User documentation

17 © Copyright 2022 Dell Inc.


Dell Customer Communication - Confidential

Exponential Growth and Adoption of SONiC

Market Differentiators Disruption Commercial Support


Innovation
Recommended
SONiC promises innovation SONiC promises innovation For mission-critical data
like Linux transformation of SONiC is predicted to be a
long-term market disruption like Linux transformation of centers, consider using
server OS Market, by server OS Market, by commercially supported
removing ties w/ hardware driver
removing ties w/ hardware distributions of SONiC to
vendors & standardizing NOS vendors & standardizing NOS ensure stability and reliability
18 © Copyright 2022 Dell Inc.
Enterprise SONiC and open-source contributions
2019 2020 1H’CY21 2H’CY21 1H’CY22 1H’CY23

Layer 3 Cloud Enterprise fabrics Edge and Telco

Managed Cloud Services


Customer Segments

FinTech Enterprise SONiC


Telco, CSP, Infrastructure • Greenfield use cases and flat
Higher Education and layer 3 network architecture
Research System
• Infrastructure as a code
Integrator
Web 2.0 Real world environment Onboarding
foundation
Enterprise features, quality,
Retail, eCommerce stability and interoperability and
Scalability
Semiconductors • Security – Certs and
Automotive Functionalities
• Edge functionality
Product Enablement • NVMe, HPC and Telco use cases

 Enterprise SONiC ProSupport  User documentation, config, and  4 HoL modules at democenter.com
deployment guides with every Rel.(6)
 Custom Deployment services  4 HoL modules for Ansible w/ E-SONiC
 Integrations (Apstra, Augtera, Ansible,
 2 courses at Dell University  40+ Channel partners trained
Fabric Design Center)
 5 Technical marketing papers  Team.Blue (Customer) case study & video
 Social media, press and analyst
testimonial
 9 Webinars coverage, industry roundtable
 Product marketing videos capturing key
 2 NSE Bootcamps  3 weeklong Sales & Services deep-dives
features, use cases, solution integrations

19 © Copyright 2022 Dell Inc.


Ecosystem for Configuration, Telemetry and Monitoring

20 © Copyright 2022 Dell Inc.


Dell Networking Products Portfolio
Products are aligned per market trends, technology evolution, and best practices

Fabric/Spine switches Top-of-rack/Leaf switches Module I/O Next-generation access

FN-IOM for
FX2 Virtual SD-WAN
Edge Edge
MX modules for MXL/IOA
PowerSwitch Z series PowerSwitch S series for
Platform Powered by
MX7000 uCPE Versa
(100G to 400G (1G to 100G) M1000e

Edge Networking software Ecosystem partners

SmartFabric OS10 SFS

Dell Enterprise
PowerSwitch E series SONiC
(1G/10G – PoE)

21 © Copyright 2022 Dell Inc.


Expanding Use Case Enablement

22 © Copyright 2022 Dell Inc.


Front-End Fabric

Enabling GenAI Use case with Dell Enterprise SONiC


Use case: Cluster / Storage / Access Fabric

Platforms:
400G/100G dense deployments
Z9664F-ON, Z9432F-ON (400G)
100G Standard deployments
Z9664F 1 Z9664F 16 S5448F-ON, S5232F-ON (100G)
25G Standard deployments
S5296F-ON, S5248F-ON
1G OOB
N3248TE-ON

Back-End Fabric Back-End Fabric


Use case: GPU to GPU Connectivity
Platform: Z9664F-ON [Tomahawk4, 64x400G + 2x 10G]
400G-SR4 Scale:
Z9664F 1 QSFP-DD Z9664F 32 • Single switch topology: Up to 64 GPUs
• Fabric topology: Up to 2048 GPUs

Roadmap 1HCY24: Z9864F [Tomahawk 5, 64x800G]


Expected Scale:
• Single switch topology: Up to 128 GPUs
• Fabric topology: Up to 8192 GPUs

Network Operating System


Enterprise SONiC Distribution by Dell Technologies
4 XE9680 4 XE9680
Release 4.2
with 8 GPUs with 8 GPUs
Feature Highlights:
• RoCEv2 – Lossless fabric, PFC, ETS, ECN
• Cut-Through Switching
• Enhanced Hashing

Front-End Fabric
Monitoring & Orchestration
Z9664F / Z9432F / S5448-ON / S52xx-ON N3248TE Augtera Monitoring solution
BeyondEdge Fabric Orchestration Solution
Connection to storage and application traffic OOB Management

23 © Copyright 2022 Dell Inc.


Delivering an Ethernet Powered infrastructure for GenAI use
cases

High Performance Compute Next Gen AI Silicon Ethernet Software Driven Innovation
and Storage Infrastructure Fabric
Validated architecture solution High Bandwidth Connectivity with High Radix Resilient, Scalable, and Open-Source
and Low Power

Accelerated AI High Performance Up to 51.2 Terabits/sec Bandwidth Cloud Inspired Architecture


Energy efficient Low Latency Cognitive Routing Open and Extensible
Flexible Architecture Secure & Protect Shared Packet Buffering Intelligent Orchestration and Monitoring
Solutions – 3rd party tools (BeyondEdge,
Augtera) and Dell internal tools

Storage Networking
Server
PowerScale, PowerSwitch
PowerEdge
ECS, ObjectScale (TH4, TH5*)

24 Copyright
© Copyright
© Dell Inc.
2022All Dell
Rights
Inc.Reserved.
*Roadmap/Scoping
Why Dell for Generative AI?
Dell's unique approach leverages simplified, tailored and
trusted solutions that combine your data with the power of
Generative AI to drive innovation and tangible business value.

Dell is the #1 worldwide provider in AI


Infrastructure

We offer full-stack scalable Generative AI


solutions with validated reference
architectures

Provide deep AI expertise at every stage


to accelerate tangible time-to-value

1 Source: IDC Semiannual AI Tracker: Worldwide Server and Storage Revenue, 2021 and 2022 H1
25 © Copyright 2022 Dell Inc.
Summary
Essential networking platform for Dell Technologies Innovation Engine
Market/Customer Trends Customer Outcomes Focus Areas
Changes in traditional markets and Optimizing enterprise solutions Focusing on software and innovation
emergence of new opportunities and investing in growth markets

Integrated, Embedded, Enabling synergy


Modernize Emerging
Edge to Core to Cloud and Attached Networking in complete
Customers seek a trusted technology partner Infrastructure the Core Growth
Providing connectivity advantages
with a seamless end-to-end solution with Dell Servers and Storage Solutions Business Areas

Control Orchestrators
Emerging Tier 2
Edge Investing in
CSP
Networking market macro
Address growing VLE / Fabric Next set of innovation
trends
Cloud-inspired, Outcome-based new use cases APEX
Telecom
Customers want to spend
less on managing technology and NOS SONiC
more toward producing results
Completing
Software Defined Networking Dell Technology Network
Focus on solution, not product Switches, DPU
value prop Hardware

Software, not Hardware Solution Enablement


Disaggregation from hardware provides
more choice, flexibility and protects against Traditional Networking Drive solutions in Storage Server Solutions
supply chain disruption End to End Dell Networking portfolio DC and Edge
APEX, Multi-cloud, Edge, Telco

26 © Copyright 2022 Dell Inc.


Thank You!
Learn more at
dell.com/networking

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy