17
1 Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables Department of Computer Science and Information Engineering National Cheng Kung University, Taiwan R.O.C. Authors: Junghwan Kim and Jinsoo Kim Publisher: ICCSA 2008 Present: Chen-Rong Chang Date: December, 24, 2008

Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

  • Upload
    berg

  • View
    64

  • Download
    0

Embed Size (px)

DESCRIPTION

Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables. Authors: Junghwan Kim and Jinsoo Kim Publisher: ICCSA 2008 Present : Chen-Rong Chang Date: December , 24, 2008. Department of Computer Science and Information Engineering - PowerPoint PPT Presentation

Citation preview

Page 1: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

1

Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Department of Computer Science and Information Engineering National Cheng Kung University, Taiwan R.O.C.

Authors: Junghwan Kim and Jinsoo Kim

Publisher: ICCSA 2008

Present: Chen-Rong Chang

Date: December, 24, 2008

Page 2: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

2

Outline

Introduction IP Address Lookup with Partitioned Prefix

Tables Update Scheduling Algorithm Experiment Environment Performance Evaluation

Page 3: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Introduction

Most of the hardware approaches for IP address lookup are based on TCAM

So TCAM can store variable-sized prefixes and find the longest matching prefix in a single cycle using the associative memory.

However, it requires the ordering of prefixes and also a priority encoder logic to find the longest matching prefix.

3

Page 4: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

IP Address Lookup with Partitioned Prefix Tables(1/3)

A conventional TCAM-based IP address lookup scheme needs both a priority encoder and some ordering constraint on prefixes.

If a prefix table is partitioned into several sets so that there can be at most one match in each set, then both the requirements can be eliminated

4

Page 5: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

IP Address Lookup with Partitioned Prefix Tables(2/3)

Since at most one match can occur in each partitioned table, we don’t need to maintain ordered prefixes for finding the best match.

It makes update operations fast because there are no movements of existing prefixes for the ordering constraint.

5

Page 6: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

IP Address Lookup with Partitioned Prefix Tables(3/3)

6

Page 7: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Update Scheduling Algorithm(1/5)

The first step of the insertion is to find the available partitioned tables to which a new prefix can be inserted. The available partitions must contain free space and be adaptable for the prefix.

The second step of the insertion is to schedule the insertion, i.e., to decide one partition among all available partitions.

7

Page 8: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Update Scheduling Algorithm(2/5)

Four update scheduling algorithms called first-fit, next-fit, random and hashing.

These algorithms focus on the minimization of updates at the OVF partition and the fairly even distribution of prefixes over the DISJ partitions.

8

Page 9: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Update Scheduling Algorithm(3/5)

9

Page 10: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Update Scheduling Algorithm(4/5)

Case1:

prefix=000010011*

Key=00001001

h(00001001)=mod(9,7)=2

10

Page 11: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Update Scheduling Algorithm(5/5)

11

Case2:

prefix=000010110*

Key=00001011

h(00001011)=mod(11,7)=4

Page 12: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Experiment Environment

12

Page 13: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Performance(1/3)

13

Page 14: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Performance(2/3)

14

Page 15: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Performance(3/3)

15

Page 16: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Conclusion(1/2)

The next-fit and the random have the characteristics of evenly distributed updates on partitions compared to other algorithms.

The first-fit has a problem that the number of updates of the OVF partition is radically increased due to the biased prefix distribution among DISJ partitions.

16

Page 17: Scheduling of Updates in an IP Forwarding Engine with Partitioned Prefix Tables

Conclusion(2/2)

The hashing algorithm is much better than any others in a point of view that there are fewer updates and prefixes in the OVF partition.

The results from a real-world routing table and an update stream show that the numbers of updates and prefixes in the OVF partition can be quite well controlled by the algorithms

17