On the Aggregatability of Router Forwarding Tables Author

advertisement

On the Aggregatability of

Router Forwarding Tables

Author :

Xin Zhao, Yaoqing Liu, Lan Wang and Beichuan Zhang

Publisher:

IEEE INFOCOM 2010

Presenter:

Li-Hsien, Hsu

Data:

9/28/2011

1

I. Introduction

 Two types of tables used by routers:

RIB(Routing Information Base) for routing

FIB(Forwarding Information Base) for forwarding

 FIB is derived from RIB. FIB usually uses high performance memory, which is more expensive and more difficult to scale. Therefore, their size is a more immediate concern to ISPs and vendors.

2

I. Introduction

Routing Scalability Problem

RIB growth => FIB growth

FIB growth: A high priority concern

(From: bgp.potaroo.net)

3

FIB Aggregation(FA)

 What is FA?

Within one router, combines multiple RIB entries with the same next hop into one.

 FA pros and cons

− Purely local no change to routing protocol

No impact on packet forwarding

Compatible with other proposed routing scalability solution(IPv6)

But extra CPU processing time

4

Forwarding Correctness

 Strong forwarding correctness

− Longest match before/after aggregation ends up with the same for all prefixes

 Weak forwarding correctness

− Prefixes with Non-NULL nexthops, the same

− Prefixes with NULL nexthops, might routable after aggregation

− extra routable space

5

FIB Aggregation Techniques &

Algorithm

Filled nodes are extra routable space introduced by the aggregation.

4A, 4B.

6

Updates Handling

Full aggregation per update is costly

 Significant computation overhead

Three approaches to handle routing changes to keep computation overhead low:

Operators choose an appropriate level of aggregation.

Incrementally update the aggregated FIB

 Minimize computation, not the table size

 Re-run full FIB aggregation periodically

 The trigger can be a timer, a threshold on FIB size, and/or current router CPU load

7

Evaluation

Data Source

− BGP routing tables and updates from RouteView

Project

Evaluation Platform and implementation

− Commodity PC, single thread process

− Algorithms implemented in C without optimization

8

Table Size after FA

RouteViews Oregon tables on 2008.12.31

Each level reduces FIB size more.

Level-1 30%~50%, Level-4 60%~90%

9

Table Size Over Time

Median of table size ratio, 2001~2008

An overall slightly decreasing trend(, suggesting that the

FIB has become more amenable to aggregation over the years.) 10

What does the ratio mean?

2006.10

2000.06

 If Level-4 applied, router deployed in 2000 can still be used today

11

Computation Time

 Computing time only takes tens to several hundreds milliseconds

12

Updates Process

D D/7,254,478 C A A/B B B/C

 Among all the updates, 2,914,020 of them cause changes to unaggregated FIB.

13

Periodical Re-Aggregation

 With threshold 150,000, on average the FIB needs to be re-aggregated every 5 days

14

Conclusion

 The table size can be reduced by 30-70%, which translates to 2-8 years extra router lifetime

 The computation overhead is small and can be controlled by incremental update handling plus periodic re-aggregation.

15

Reference

 www.cs.arizona.edu/~zhaox/slides/FIB-Aggregation-INFOCOM2010.ppt

16

Download