A Framework for Scalable Global IP-Anycast

Sigcomm 2000, Dina Katabi

Presented by Wei Yu

Outline Motivation Introduction Background work Idea & design detail Performance Evaluation Result

Motivation A scalable architecture for Global IP-

Anycast Existing design drawback

– Global Distribute route to individual group =>not scalable

Idea: dividing inter-domain routing into two parts: – Build inexpensive anycast route, consume no

BW and storage space– Edge domain generate anycast routes according

to the interest.

What is Anycast?

Anycast ( RFC 1546): sender to access the nearest of a group of receiver sharing anycast address

Application: service discovery & host autoconfiguration

EX: replicated FTP server, mirror web server

Background & related work IP level solution: proposed by IPv6

• anycast routing

Application level solution: Need directory system – disadvantage:

• overhead to collection information• lack bootstrap mechanism

– advantage:• easy to deploy

Difficulty: scalable architecture– defies hierarchical aggregation– routing table

Idea Unicast routing paradigm problem

– waste routing resources– ex: US & Lindon => devote more resource

to frequency access anycast group Using cache technology Using edge domain resources, why?

– Prevent wasting routing resources– push most of work to edge domain

Edge domain discover, store and maintain anycast route

Design detail

Address architecture

Design Detail Anycast group classification

Design Detail Routing internal anycast group:

– using unicast routing protocol

Routing unpopular anycast group– using default route to forward packets

Routing popular anycast group– initial search, receiving search, receiving reply, search

scope: TTL

Design details

Learning routes to popular anycast group

Performance Evaluation

Main results– nearest route – overhead ( millions global group)

Supposition– provide lower bound on efficiency– upper bound of overhead

Success route performance

Overhead

Scalable

Question ?