CS5223

Distributed Systems

Lecture 4: Naming

Instructor: YU Haifeng
 Today’s Roadmap
 Chapter 5 of textbook
 P2P and DHT is a important topic and will be covered in a
separate lecture

 Basic concepts in naming

 Simple naming mechanisms

 Domain name service

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 2
 Motivation for Naming
 Example: We want to visit www.nus.edu.sg
 The web server is located in some room that we don’t know
 There may be multiple web servers
 ISP’s may want to serve the content from their own
machines – Akamai
 www.nus.edu.sg  ip address  some physical machine
(with specific MAC address)

 Example: Distributed file system

 We know the file name (and full path), need to find out
where the file is (i.e., on which machine, which disk sector,
etc)
 The file may migrate from one machine to another without
changing its name

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 3
 Basic Concepts in Naming
 General definition for naming:
 We have a “name” and we want to find/locate the “entity”
 Definitions in the textbook is unnecessarily convoluted, and
sometimes circular – please ignore those
 A lot of things can be called “naming” – just a definition issue – no
need to draw a black-and-white boundary

 Name space: The set of all possible names in the given context

 Naming system: The system that enables you to find the “entity”
 Name resolution
 The naming system can be either non-distributed or distributed
 Same tradeoff as before…

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 4
 Different Kinds of Names
 Every name is a bit string
 Some are human-friendly

 Hierarchical versus flat

 Alias
 Multiple names for the same entity
 Similar as symbolic links

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 5
 Merging Name Spaces
 We have two different name spaces
 Now we want to define a unified name space
 Name space merging
 Generalizes to multiple name spaces…how?

 Example: Merging two network file systems

 Example: Two users want to give each other access

to files in their home directories

 Main issue to take care of: There can be name

collisions

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 6
 Solution 1: New Root Node
 Merge by adding a new root node and make existing
root nodes its children
 In principle, you always have to start in the new root
 But you can avoid this by including the identifier of node
from where resolution should start
 Example, ~/XXX

 Problem: existing names need to be changed

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 7
 Example: New Root Node

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 8
 Solution 2: Mounting
 Introduce nodes that contain the name of a node in a
“foreign” name space

 Mount point: (Directory) node in naming graph that

refers to other naming graph

 Mounting point: (Directory) node in other naming

graph that is referred to

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 9
 Example of Mounting

Mounting
Point

Mount Point

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 10
 Simple Naming: Broadcast
 The “naming system” may or may not be a stand-
alone software module

 Broadcast to resolve a name

 Used in early day of computing
 Simple is good!

 Drawback: performance
 Alleviated but not fully addressed by using multicast

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 11
 Simple Naming: Forwarding Pointers
 Forwarding pointers: Same idea as mail forwarding

 The machine A with ip_addr1 was initially the web server for
www.nus.edu.sg
 We want to use another B machine now with ip_addr2 as the server

 A will forward the HTTP requests to B

 Or A can inform the client of B’s address

 Drawback:
 A needs to stay there forever
 The forwarding chain can become long

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 12
 Simple Naming: Home-Based Approach
 Each entity has a “home”
 The entity will register its IP address (which may change)
with its home

 To find the entity, we contact the home first

 Only minor difference from forwarding pointers

 Maximum one-hop forwarding

 Drawback:
 The home needs to be there forever (i.e., have a fixed IP
address)
Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 13
 Example: Home-Based Approach

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 14
 Domain Name System: DNS
 For looking-up IP addresses
 Example: what ip address www.nus.edu.sg corresponds to

 DNS dates back to over 30 years ago

 Replaces the original Internet naming scheme that keep everything
in a master file (lack of scalability and freedom for each domain to
assign internal names)

 Overall, DNS has been a very successful distributed system

 Many of the early design decisions turn out to be critical later
 Additional reading (non-compulsory): “Signposts in cyberspace:
The Domain Name Systsme and Internet Navigation” by R. Levien,
National Academic Research Council, 2005
 We will study it as a case study…

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 15
 Side-track: The Role of Case Studies
 RMI: Case study for RPC
 DNS: Case study for naming system

 Distributed systems has two aspects

 Algorithms: Science – can rigorously prove things, can reason about
properties mathematically
 End-to-end systems: Art – too complex to capture mathematically
 How to do better art
 Subjective judgment based on one’s knowledge and experience
 Experience comes from case studies
 It is easy to miss some aspects when thinking about designs on paper

 This has a strong analogy to law and history

 In certain law systems, arguments are made by drawing connection to
historical law suits

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 16
 DNS Name Space

DNS name (max length 255)

www.comp.nus.edu.sg/~yuhf
label (max length 63)

 Top-level domains
 com, edu, net, org, etc.
 sg, cn, fr, etc.

 The machine corresponding to XXX.XXX.sg may or may not be

in Singapore

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 17
 The DNS Tree

(from http://www.rhyshaden.com/dns.htm)

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 18
 Relative Names in DNS
 Not supported in DNS specification

 However:
 Your local software may automatically append the omitted
part
 Example, you can type ssh suna on any SoC unix machine,
and you will connect to suna.comp.nus.edu.sg
 If suna.comp.nus.edu.sg does not exist, ssh will try to
connect to “suna” (which does not exist either)
 Sometimes the local software may try appending different
suffixes

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 19
 Domains in DNS
 A domain is a name space for which there is a single overall
administrative authority to assignment the DNS names
 Example, the collection of all XXX.nus.edu.sg is a domain

 A domain’s name is the common suffix of all the DNS names in

the collection
 Example, nus.edu.sg is the domain’s name
 By definition, a domain’s name is also a DNS name

 Potential confusion
 A DNS name is sometime called domain name, which is different
from domain’s name defined above
 We will only use DNS name and domain’s name to avoid confusion

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 20
 Resolving DNS Names:
The Perspectives of a Client
 Web browser
 Make query to DNS server (how to find the DNS server?) to
resolve www.nus.edu.sg into an IP address
 Send HTTP request to that IP address at port 80

 ftp, telnet, ssh, smtp(email) are all similar

 Reverse resolution:
 Translate an IP address into a DNS name
 Only for IP addresses within the local domain

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 21
 Resolving a DNS Request
 The client needs to know the “local DNS server”
 The IP address of this server is given to you by your ISP
 Bootstrapping

 Client sends DNS query to local DNS server S

 Containing the DNS name e.g., www.nus.edu.sg

 S checks whether it has either authoritative RR or

cached RR for
 www.nus.edu.sg
 nus.edu.sg
 sg

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 22
 Root DNS servers
 If no, S will send quest to the root DNS server
 How does S know where to find the root DNS server?

 13 root servers with well-known IP addresses

 https://www.iana.org/domains/root/servers

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 23
 Resolving a DNS Request
 The root DNS server sends back to S the IP address of the DNS
server responsible for the domain “sg”

 S contact that DNS server

 And then recursively the DNS server for “edu.sg”, “nus.edu.sg”, …

 After a DNS name is resolved, the RR will be cached

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 24
 Critical Design Decisions in DNS
 Delegation of responsibility
 Allow individual domains to assignment their own DNS
names
 Hierarchical naming strcuture

 Caching and replication

 Controls the overhead of DNS lookups
 Controls the load on the root DNS servers

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 25
 Vulnerability of DNS
 The root servers – “single points” of failure
 Target of DoS attacks
 Already happened

 Alternatives:
 Completely decentralized distributed naming system for the
Internet
 Already possible technically, but backward-compatibility
issues need to be addressed

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 26
 History Readings (Non-compulsory)
 DNS root servers under attack
 http://en.wikipedia.org/wiki/DNS_Backbone_DDoS_Attacks

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 27
 The Fundamental Need
for Bootstrapping in Naming
 To resolve a name, you need to know at least one
bootstrapping point
 Broadcast: Bootstrapping point not needed (or every
machine is a bootstrapping point?)
 Forwarding pointers: The first machine you contact
 Home-base approach: The home
 DNS: Your local DNS server and the root DNS servers

 Called “closure mechanism” on the textbook

 Bootstrapping mechanism significantly influences the

“goodness” of a naming system

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 28
 Today’s Summary

 Basic concepts in naming

 Simple naming mechanisms

 Domain name service

Haifeng Yu, CS5223, Some Contents Adapted (with permission) from © R.Ayani, G.Tan 29