QoSAware Path Protection in MPLS Networks

1
QoS-Aware Path Protectionin MPLS Networks
Satish Tripathi University of California at
Riverside

• Ashish GuptaAshish GuptaBijendra Jain
• Indian Institute of Technology Delhi

2
Overview

• MPLS networks
• Special need for path protection
• Approaches to path protection in MPLS networks
• Link, node based
• LSP based
• Segment Based Approach
• Mechanisms (detection, notification and path
  switching)
• Algorithm for segment identification
• Some simulation results

3
Path protection
1
47.1
3
3
1
2
1
2
47.3
47.2
3
2

• Routing algorithms for IP networks take seconds
  to re-compute routes
• Voice video are sensitive to switch-over time
• require switch over in less than 50 to 100 ms
• Identify and set-up back-up paths a-priori

4
(No Transcript)
5
Path protection in MPLS networks

• Protection against link and node failure
• We consider both types of failures
• Single vs. multiple failures
• We largely consider single failures, but
• Centralized vs. distributed computation and setup
• We permit centralized computation

6
Major considerations

• Bound switch-over time (and consequent loss of
  data)
• Of the order of 50 ms to 150 ms
• Meet SLA commitments
• viz. QoS constraints
• Path availability
• End-to-end delay
• Jitter
• Packet drop rate
• Use resources for path protection efficiently

7
Path protection in MPLS networks

• Complete LSP by-pass
• allocation of resources along back-up path is
  efficient
• time to detect failure and switch over is large
• difficult to identify a node-disjoint path that
  also meets the specified QoS constraints

8
Path protection in MPLS networks

• Link by-pass
• allocation of resources for back-up paths is
  unlikely to be efficient
• fault detection and switch over can be fast
• cannot be sure about QoS resulting from any
  failure
• does not address node failure

9
Path protection in MPLS networks

• Node by-pass
• allocation of resources for back-up paths is
  unlikely to be efficient
• fault detection and switch over can be very fast
• cannot be sure about QoS resulting from any
  failure

10
Segment Based Protection

• The Main Idea
• Look at the path as a sequence of segments and
  protect each segment separately

11
Segment based protection a proposal

• flexibility in identifying segments
• schemes to protect LSP, links, or nodes are
  special cases
• efficient allocation of resources for back-up
  paths
• bounds on fault detection and switch over time
• ability to identify back-up paths that meet
  specified QoS constraints

12
Focus of our paper

• develop algorithms to identify segments, and
  back-up paths, such that
• switch over time (time for which packets are lost
  between failure and recovery) is bounded
• path resulting from any single failure continues
  to satisfy given QoS constraints
• resources are used efficiently (or more
  precisely, the number of segments is minimized)

13
Algorithm for Bounded Switch over time
Switch over time The time for which packets are
lost between failure and recovery
14
Fault Detection , Location and Notification

• Faults detected using live-ness messages with
  periodicity Ttest
• Notification messages to segment switching
  routers (SSR)

15
Analysis
R5 fails at time t
After t - OWD(R2,R5), packets uncertain
R2 gets notified at t ttest OWD(R5,R2)
16
Analysis
Bound on time during which packets are lost RTT(
Ri , Rj ) Ttest
17
Identifying Segments A greedy algorithm

• Identify segments such that
• switch-over delay is bounded (for instance, 60
  ms)
• Fewest no of segments
• Example computation
• 11 hop LSP, Ttest 10 ms

R11
R10
R9
R8
R7
R6
R5
19
11
13
14
11
10
8
3
10
15
10
18
Another algorithm to identify segments

• Example consider network with link RTT 10 ms,
  Ttest 5 ms
• bounded switch-over delay of 40 ms
• with as few segments as possible
• disjoint, loop-free back-up paths exist

10 ms
10 ms
10 ms
10 ms
10 ms
Ingress router
Segment switch router
19
Other algorithms to Identify Segments

• Algorithms to identify segments such that
• Switch-over time is bounded
• Fewest no. of segments
• disjoint, loop-free back-up paths exist
• QoS constraints are satisfied in case of ONE
  failure
• End-to-end delay, Jitter, Drop rate

20
Identifying Back-up paths

• An individual back-up path
• is node- and link-disjoint from the segment it
  protects
• may terminate at any node beyond the last node on
  the segment
• Gives greater flexibility
• The back-up paths taken as a whole
• Must not form loops (equivalently, individual
  segments must be node- and link-disjoint from all
  nodes, links in earlier segments)
• Together with the original path, the back-up
  paths must satisfy QoS constraints assuming at
  most one link/node failure

21
End to End Delay
22
End to End Delay
Finding the backup path
d1
d2
d3
d2 d3
d3
0
d1 d2 d3

Dummy node
- Can use shortest-path approach to find the
backup path - Backup path can land at multiple
nodes
23
Other work in the paper
24
Description of Simulation Setup

• An MPLS network with
• 50 Nodes
• 82 Edges
• Random LSP that require 20 to 70 units of BW
• RTT of each link 8 to 12 ms
• BW between 3000 and 10000 units
• Periodicity of liveness messages 2 ms
• BW 50 to 100
• Results indicate advantages of segment based
  approach

25
Description of Simulation Setup

• Topology used

26
Simulation Results

• BW reserved for back up vs. number of LSP for
  different bound on switch-over time

27
Simulation Results

• Reserved BW vs. switch-over time

28
Summary

• Segment based approach offers a range of schemes
  for path protection
• From link or nodes to segments, to paths
• The approach permits one to insist that back-up
  paths continue to provide committed QoS even when
  there is a failure
• The approach ensures that resources are reserved
  only to the extent necessary
• Many of our algorithms are good, and provably
  correct, but may not be optimal