Enhancing the Internet Survivability using IP Fast Rerouting Kang - - PowerPoint PPT Presentation

1

Enhancing the Internet Survivability using IP Fast Rerouting

Kang Xi and H. Jonathan Chao

• Dept. of Electrical & Computer Engineering

Polytechnic University {kang,chao}@poly.edu

2

Basics of IP Routing

Find the src/dst paths for packet forwarding Paths to the same dst form a tree

4 1 9 2 7 5 8 6 3 4 1 9 2 7 5 8 6 3

3

What if a failure happens?

Failures are common

in networks

A failure may affect

several paths

Traditional Recovery:

Advertise the failure Recalculate all the

paths

Problem: Slow

4 1 9 2 7 5 8 6 3 4 1 9 2 7 5 8 6 3

4

Failures in Sprint Backbone (2002)

5

What is IP fast re-routing?

Calculate backup

paths before failures

• ccur

Use backup paths

immediately after a failure is detected

Advantage: fast

recovery, highly applicable

6

Challenges

How to find loop-free

backup paths?

How to coordinate

routers during the recovery?

7

Use the Disjoint Paths in Circuit Switching?

Only ONE entry/output port in the routing table for the destination.

8

Existing Work

Relatively new topic IETF, Cisco, Alcatel are involved in the

research

Only a few papers published Focus on single failures

9

Our Progress

A set of schemes called Efficient SCan for

Alternate Paths (ESCAP)

A simple solution for single-link/node failures A solution to handle double-link failures No failure advertising required

10

Single-Node Failure Recovery

Example: Failure Scenario

5 1 15 2 10 6 16 7 3 14 13 12 11 8 4 9

Example: Search Procedure

Example: Search Procedure

Example: Result

Algorithm

Denote a tree/sub-tree with T(k), where k is the

index of the root.

Use depth-first search, for each node k (except the

root), do the following:

Introduce a failure to node k Node k has mk children, denoted as c1, c2, … cmk. Need

to find an exit for each of T(c1), T(c2), … T(cmk)

Dye the nodes in T(k), all the other nodes are white Repeat the following two steps until all node are white

Use breadth-first search on T(k), for each node, check if it has a

white neighbor

The first hit found an exit for one of T(c1), T(c2), … T(cmk), dye

that sub-tree white

Forwarding Policy

4 1 9 2 7 5 8 6 3

Performance Evaluation

Performance Metrics

Path lengths Traffic load on each link

Method

Topology

Practical topologies: COST239, NSFNet

Service Demand

1 Mb/s between any two nodes

Comparison

IPFRR Shortest path: Normal Shortest path route recalculation under failure

Topology

Path Length Distribution

Explore all possible failures, observe the distribution of

the path lengths

Load on Each Link

24

Conclusions

Three schemes: single-link, single-link/node, double-link

failure recovery, respectively

100% failure coverage Proven correctness Practical to today’s IP networks Support of multi-path routing. Future Research

Multi-failure recovery Failure recovery in inter-domain routing Survivable topology design Building a testbed