Ashish Gupta Ashish Gupta

1
Ashish GuptaAshish Gupta
Under Guidance ofProf. B.N. Jain
Department of Computer Science and Engineering
Advanced Networking Laboratory
2
Overview

• Current routing in the Network
• Why MPLS ? Like ATM over IP
• The need for fault tolerance today
• What is the problem with current Path Protection
  Schemes Give no assurance for QoS
• The challenge give us QoS guarantees but dont
  use too much bw
• What we have proposed SBPP
• The need for algorithms for various QoS
  constraints
• Present the algorithms in a simple manner only
  the switch over time and end to end delay
• Mention other work done how we finally solved
  the reliability problem
• We have made a Admission Control simulator in
  C the algo performs really well ! Gives you
  tradeoffs now for various QoS Constraints
• The visualization of our AlgorithmsThe need and
  its features, and the demo
• Conclusion no worries on failure now
• Over all work done (besides the algos) , we have
  also written a paper to be send to ICCC
• Any questions ?

3
Routing - The IP Way
All packets follow the same path
Need for more control on routing
4
Multiprotocol Label Switching
1
47.1
3
3
2
1
1
2
47.3
3
47.2
2
Gives explicit control over routing according to
traffic requirements
Concept of Label Switching and Label Switched
Paths (LSPs)
5
Need for Fault Tolerance
1
47.1
3
3
1
2
1
2
47.3
3
47.2
2
Modern delivery services need to be robust...
6
Method of Path Protection
1
47.1
3
3
1
2
1
2
47.3
3
47.2
2
Enables faster recovery than Layer 3 Mechanisms
Using Dynamic Routing Algorithms, recovery may
take order of seconds !
Reserve an alternate path and use it in case of
failure
7
Major Concerns for Path Protection
Should provide quick recovery after a failure
Optimal Use of Backup Resources
8
Global Path Protection
Primary Path
9
Local Path Protection
10
What's Missing ?

• The solutions are fixed
• No assurance for various QoS constraints

We need guarantees that in case of failure, we
will not lose packets for more than 50 ms
Or even in case of failure, the end to end delay
will not exceed 150 ms
Fault Tolerance with QoS Guarantees
11
Segment Based Approach
to Path Protection
12
Segment Based Path Protection

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

• Results in fewer backup paths conserves
  resources
• Can meet QoS constraints in a tight manner
• Gives flexibility How ?

13
An Example

• A Segment is a set of nodes which consists of
• Segment Switch Router protects the segment
• Protected Nodes

How to divide the path into segments such that
? If any segment fails, specified QoS constraints
can be guaranteed
14
Algorithms for segmenting the path

• Various QoS constraints considered
• Bounded Switch-Over Time
• End-to-End Delay
• Jitter
• Reliability
• Combination of above

15
Algorithms for bounded Switch over time
Switch over time The time for which the packets
are lost between the failure and recovery
Why do we lose packets ?
16
Analysis for Bounded Switch over time
Using Timing Analysis, we derived a limit to the
possible segment size
RTT( Ri , Rj ) Ttest lt ?
17
The Algorithm
Divide the path into minimum number of segments
such that the inequality is satisfied for each
segment

• A new Adaptive Algorithm for Bounded Switch
  over time has been developed
• Takes into account the topology and bandwidth
  reservation in the network

• Intuitive choice Greedy Algorithm
• Problems with Greedy Algorithm

18
The Adaptive Algorithm

• Assures QoS satisfaction
• Proved to be optimal
• Robust
• We have implemented it using the LSP Admission
  Control Simulator and tested it
• Results indicate its efficiency

19
Algorithm for End-to-End delay
An important parameter in multimedia
communications
20
Analysis for End-to-End delay
21
The Algorithm
We need to satisfy the inequality for each
possible path

• For each segment (found using previous
  algorithm), find the shortest possible backup
  path
• Will give the minimum value of end-to-end delay
• Check if inequality is satisfied, otherwise
  reject the current segment

22
More QoS constraints

• Jitter similar to End-to-End delay
• Multiple QoS constraints
• Bounded Switch Over time
• End-to-End Delay
• Jitter
• Developed an algorithm using Dynamic
  Programming Approach

23
Algorithm Developed for Reliability

• Finding Reliability between two nodes An NP
  complete problem
• Earlier we were engaged in finding bounds
• But turns out, Segment Based Approach allows
  for simpler analysis !

24
Algorithm Developed for Reliability

• An O(No. of Links (No. of Segments)2 )
  Algorithm to find exact path reliability !
• Algorithm for finding the most reliable backup
  path
• A heuristic Algorithm has been developed for
  meeting bounds on Reliability for the protection
  configuration

25
(No Transcript)
26
The Visualization System

• A visualization system developed for the
  Segment based Algorithms based on POLKA
• Closely Integrated with LSP Admission Control
  Simulator
• Visualizations are Dynamic not Static
• Aids in understanding how the algorithms work
• Assists in establishing correctness of
  algorithms and simulations

27
Topology Used
28
Visualization Demonstration !
29
Experimental Results
Aim of the Experiments To evaluate the resource
related advantages of Segment based Approach,
while assuring QoS constraints
30
Experimental Results

• An LSP Admission Control Simulator has been
  developed in C
• Simulates multiple networks with different
  reservation policies and compares results
• We implement the Segment Based Algorithm for
  Bounded Switch over time for providing Protection
  to LSPs generated
• BW Sharing Mechanisms also implemented

31
Experimental Results

• Simulation Setup
• Topology with 50 routers and 82 links
• LSPs setup using Djiktras Algorithm
• BW of each link 3000 to 10000 units
• Delay of each link 8 to 12 ms
• BW requirements of each LSP 20 to 70 units

LSP requests were generated with different QoS
parameters
32
Comparison of Different Schemes
33
Variation of Switch over time bound
34
Rejection Rate vs Bound on Switch over time
35
Conclusion

• Today major networking companies are quickly
  shifting towards MPLS-based networks
• Major Players Cisco , ATT, Level 3, Hewlett
  Packard, UUNet, MCI World Com
• Within an MPLS-enabled network, even packet
  voice services such as voice over IP will benefit
  from reduced latency and congestion control
• Ranjeet Sudan, MPLS product manager at Cisco

Efficient Fault Tolerance in such services is a
major concern
36
Conclusion

• For providing reliability to such services we now
  have Fault Tolerance Schemes which
• Guarantee satisfaction of various QoS constraints
  in case of failure
• Use backup resources in an optimal manner

37
Overview of BTP

• Mechanisms for Detection and Notification
• Algorithms for various QoS constraints
• Bounded Switch over time
• End-to-End delay
• Jitter
• Combination of above
• Reliability
• Issues relating to backup path Sharing
• Admission Control Simulator developed for
  above
• Visualization System

38
Paper being submitted to
Being held in Mumbai this year
39
Thank You