Title: A Study on the Efficacy of Regular Virtual Topology Design Heuristics for Optical Packet Switching
1- A Study on the Efficacy of Regular Virtual
Topology Design Heuristics for Optical Packet
Switching - Olufemi Komolafe - University of Strathclyde,
Glasgow, UK - David Harle - University of Strathclyde, Glasgow,
UK - David Cotter - Corning Research Centre, Ipswich,
UK
2 REGULAR VIRTUAL TOPOLOGY DESIGN
Cost
INPUTS
SOLUTIONS
Optimisation techniques
3 PROBLEM INPUTS
Cost
INPUTS
SOLUTIONS
Optimisation techniques
4Exemplar Regular Virtual Topology
Manhattan Street Network
- Clockwork Routing
- Simple packet routing
- No optical contentions
- Favourable performance
- No resequencing _at_ destinations
5Arbitrary Physical Networks
6Arbitrary Physical Networks
7Arbitrary Physical Networks
8Arbitrary Physical Networks
9Arbitrary Physical Networks
10Arbitrary Physical Networks
- Unique if different values for any of
- maximum degree
- degree variance
- mean inter-nodal distance
- maximum inter-nodal distance
- diameter
- inter-nodal distance variance
- number of bridges
- link whose removal disconnects network
11 COST
Cost
INPUTS
SOLUTIONS
Optimisation techniques
12Approach
Map MSN nodes onto physical topology
MSN
Physical topology
13Approach
Establish lightpaths between nodes
MSN
Physical topology
14Numerous different mappings
15(No Transcript)
16(No Transcript)
17(No Transcript)
18(No Transcript)
19(No Transcript)
20(No Transcript)
21(No Transcript)
22Cost
- Mean lightpath length
- impacts number of hops packets traverse
- affects number of ? needed
- indicates number of optical cross-connects
traversed between adjacent MSN nodes - affects deployment of optical amplifiers
consumption of other network resources
23 OPTIMISATION TECHNIQUES
Cost
INPUTS
SOLUTIONS
Optimisation techniques
24Optimisation Techniques
- NP-hard problem
- Use heuristics to find (near) optimal solutions
expeditiously - Use heuristics that work fundamentally
differently - confidence in results and trends
25Optimisation Techniques
- Simulated annealing
- modelled on cooling of molecules to form crystal
- uphill moves allowed with decreasing probability
- Genetic algorithms
- modelled on natural evolution
- 2 different implementations
- Cycle crossover (CX)
- Partially mapped crossover (PMX)
- Hill climbing
- Random search
26 REGULAR VIRTUAL TOPOLOGY DESIGN
Cost
INPUTS
SOLUTIONS
Optimisation techniques
27 REGULAR VIRTUAL TOPOLOGY DESIGN
Cost
INPUTS
SOLUTIONS
Optimisation techniques
28 REGULAR VIRTUAL TOPOLOGY DESIGN
Cost
INPUTS
SOLUTIONS
Optimisation techniques
29 REGULAR VIRTUAL TOPOLOGY DESIGN
Cost
INPUTS
SOLUTIONS
Optimisation techniques
30 REGULAR VIRTUAL TOPOLOGY DESIGN
Cost
INPUTS
SOLUTIONS
Optimisation techniques
31 PEERING INTO BLACK BOX
Cost
INPUTS
SOLUTIONS
Optimisation techniques
32Evaluation of Heuristics Efficacy
- 2 key metrics
- Quality of final solution
- final mean lightpath length obtained
- Efficiency associated with obtaining final
solution - number of solutions considered
- corresponds to area of search space explored
- indicative of time/computational effort
- generic and portable metric
- Monitor evolution in cost as optimisation
heuristic progresses
33 Experimental Approach
- Generate randomly connected physical networks
- mean degree 2, 5, 8
- Use heuristics to deploy MSN in each network
- monitor evolution in mean lightpath length
- Find average performance for each heuristic
34Sample Results SA (Mean degree 2)
Each line shows progress of SA when deploying MSN
in unique physical topology
35Sample Results SA (Mean degree 2)
Each line shows progress of SA when deploying MSN
in unique physical topology
Mean 3.345 STD 0.238 95 CI 0.079
Mean 5.721 STD 0.707 95 CI 0.234
36Randomly Connected Networks (Mean degree 2)
37Randomly Connected Networks (Mean degree 5)
38Randomly Connected Networks (Mean degree 8)
39General Observations
Significant reduction on initial cost
Converge to similar value
40Random Search
Worst performance
41Hill Climbing
Relatively good solutions
Short convergence time
42GA - Cycle Crossover
Relatively good solutions
Short convergence time
43GA - Partially Mapped Crossover
Very good solutions
Longer convergence time
44Simulated Annealing
Uphill moves
Longest convergence time
Best solutions
45Nature of Cost Function
- Determines performance of optimisation techniques
- Catch-22 scenario
- Must apply certain optimisation techniques to
determine nature of cost function - What insight do results provide about nature of
cost function?
46Nature of Cost Function
Plateau with fissures?
47Meritocratic Ordering
Quality of Final Solution Simulated
Annealing Partially Mapped Crossover Hill
Climbing Cycle Crossover Random Search
of Solutions Considered Cycle Crossover Hill
Climbing Random Search Partially Mapped
Crossover Simulated Annealing
48Meritocratic Ordering
Quality of Final Solution Simulated
Annealing Partially Mapped Crossover Hill
Climbing Cycle Crossover Random Search
of Solutions Considered Cycle Crossover Hill
Climbing Random Search Partially Mapped
Crossover Simulated Annealing
49THANKS