impact of problem centralization on distributed constraint optimization algorithms john p. davin and...
Post on 22-Dec-2015
215 views
TRANSCRIPT
![Page 1: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/1.jpg)
Impact of Problem Centralization on Distributed Constraint Optimization Algorithms
John P. Davin and Pragnesh Jay ModiCarnegie Mellon University
School of Computer Science{jdavin, pmodi}@cs.cmu.edu
Autonomous Agents and Multiagent Systems 2005
July 29, 2005
![Page 2: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/2.jpg)
DCOP• DCOP - Distributed Constraint Optimization Problem
– Provides a model for many multi-agent optimization problems (scheduling, sensor nets, military planning).
– More expressive than Distributed Constraint Satisfaction.– Computationally challenging (NP Complete).
![Page 3: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/3.jpg)
Defining DCOP Centralization
• Definition: Centralization – aggregating information about the problem into a single agent, where:– this information was initially distributed among multiple agents, and– this aggregation results in a larger local search space.
For example, constraints on external variables canbe centralized.
![Page 4: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/4.jpg)
Motivation
• Two recent DCOP algorithms - Adopt and OptAPO:– Adopt does no centralization.– OptAPO does partial centralization.
Prior work [Mailler & Lesser, AAMAS 2004] has shown that OptAPO completes in fewer cycles than Adopt for graph coloring problems at density 2n and 3n.
But, cycles do not capture performance differences caused by different levels of centralization.
![Page 5: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/5.jpg)
Key Questions in this Talk
How do we measure performance of DCOP algorithms that differ in their level of centralization?
How do Adopt and OptAPO compare when we use such a measure?
![Page 6: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/6.jpg)
Distributed Constraint Optimization Problems (DCOP)
• Agents: A = {A1, A2 … AN}
• Variables: V = {x1, x2, … xn}
• Domains: D = {D1, D2, … Dn}
• Cost functions: f = {f1, … fk}
• Objective function F:
Goal: Find values for variables that minimize the sum cost over all cost functions (constraints).
Vxx
jiij
ji
ddfF,
),(
![Page 7: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/7.jpg)
DCOP Algorithms
• ADOPT [Modi et al., AIJ 2005]– Agents communicate variable values and
costs (lower bounds, upper bounds).– Each agent’s search space remains constant.
• OptAPO – Optimal Asynchronous Partial Overlay. [Mailler & Lesser, AAMAS 2004]– cooperative mediation – a mediator agent
collects constraints for a subset of the problem, applies centralized search.
– Mediator’s search space grows.
![Page 8: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/8.jpg)
ExampleA={A1,A2,A3}, D={0,1}, Constraints={A1!=A2, A2!=A3, A1!=A3}Adopt:
![Page 9: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/9.jpg)
ExampleOptAPO: A={A1,A2,A3}, D={0,1}, Constraints={A1!=A2, A2!=A3, A1!=A3}
![Page 10: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/10.jpg)
Key Questions
How do we measure performance of DCOP algorithms that differ in their level of centralization?
How do Adopt and OptAPO compare when we use such a measure?
![Page 11: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/11.jpg)
Previous Metric: Cycles
• Cycle = one unit of algorithm progress in which all agents process incoming messages, perform computation, and send outgoing messages.– Independent of machine speed, network
conditions, etc. – Used in prior work [Yokoo et al., 1998],[Mailler
et al., 2004]
![Page 12: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/12.jpg)
Previous Metric: Constraint Checks• Constraint check = the act of evaluating a constraint
between N variables.– Standard measure of computation used in centralized
algorithms.• Concurrent constraint checks (CCC) = maximum
constraint checks from the agents during a cycle.– Used in prior work for distributed algorithms [Meisels et al., 2002].
![Page 13: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/13.jpg)
Problems with Previous Metrics
• Cycles do not measure computational time (they don’t reflect the length of a cycle).
• Constraint checks alone do not measure communication.
We need a metric that measures both communication and computation time.
We introduce a new metric, Cycle-Based Runtime (CBR), to address this.
![Page 14: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/14.jpg)
Cycle-Based Runtime
•L =• T = time of one constraint check.
- Let T=1, since constraint checks are the shortest operation that we are interested in.
•L = communication latency (time to communicate in each cycle).
Let L be defined in terms of T
• L=10 indicates communication is 10 times slower than a constraint check.
L can be varied based on the communication medium.
• Eg., L=1, 10, 100, 1000.
![Page 15: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/15.jpg)
CBR: Cycle-Based Runtime
),(max)(0
kxccmccc i
m
kVxi
Define ccc(m) as the total constraint checks:
tmcccmLmCBRcyclesmofruntimetotal )()(
![Page 16: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/16.jpg)
Results• Tested on graph coloring problems, |D|=3 (3-coloring).• # Variables = 8, 12, 16, 20, with link density = 2n or 3n.
• 50 randomly generated problems for each size.
Cycles: CCC:
OptAPO takes fewer cycles, but more constraint checks.
0
10000
20000
30000
40000
50000
60000
70000
80000
90000
8 12 16 20
Variables
CC
C
Adopt
OptAPO
0
1000
2000
3000
4000
5000
8 12 16 20
Variables
Cyc
les
Adopt
OptAPO
![Page 17: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/17.jpg)
Density 2tmcccmLmCBR )()(
L=1
0
20000
40000
60000
80000
100000
8 12 16 20
# vars
Adopt
OptAPO
L=10
0
20000
40000
60000
80000
100000
8 12 16 20
# vars
Adopt
OptAPO
How do Adopt and OptAPO compare using CBR?
L=100
0
20000
40000
60000
80000
100000
8 12 16 20
# vars
Adopt
OptAPO
L=1000
0
50000
100000
150000
200000
250000
8 12 16 20 24
# vars
Adopt
OptAPO
![Page 18: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/18.jpg)
L=10
0
3000000
6000000
9000000
8 12 16 20
# vars
Adopt
OptAPO
L=100
0
3000000
6000000
9000000
8 12 16 20
# vars
Adopt
OptAPO
L=1000
0
10000000
20000000
30000000
40000000
8 12 16 20 24
# vars
Adopt
OptAPO
How do Adopt and OptAPO compare using CBR?Density 3tmcccmLmCBR )()(
For L values < 1000, Adopt has a lower CBR than OptAPO. OptAPO’s high number of constraint checks outweigh its lower number of cycles.
L=1
0
3000000
6000000
9000000
8 12 16 20# vars
Adopt
OptAPO
![Page 19: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/19.jpg)
How much centralization occurs in OptAPO?
OptAPO sometimes centralizes all of the problem.
![Page 20: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/20.jpg)
How does the distribution of computation differ?
• We measure the distribution of computation during a cycle as:
This is the ratio of the maximum computing agent to the total computation during a cycle. – A value of 1.0 indicates one agent did all the computation.
– Lower values indicate more evenly distributed load.
Agentsxi
iAgentsx
i
i
kxcc
kxcckload
),(
),(max)(
![Page 21: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/21.jpg)
How does the distribution of computation differ?
• Load was measured during the execution of one representative graph coloring problem with 8 variables, density 2:
OptAPO has varying load, because one agent (the mediator) does all of the search within each cycle. Adopt’s load is evenly balanced.
![Page 22: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/22.jpg)
Communication Tradeoffs of Centralization
Centralization performs best relative to non-centralized approaches at higher communication latencies.
• At high density, centralized Branch & Bound outperforms
OptAPO.
• How do the algorithms perform under a range of communication latencies?
Density=2
0
5 0 0 0 0
1 0 0 0 0 0
1 5 0 0 0 0
2 0 0 0 0 0
2 5 0 0 0 0
1 10 100 1000
L
CB
R
AdoptOptAPOCentralized
Density=3
0
1 0 0 0 0 0 0
2 0 0 0 0 0 0
3 0 0 0 0 0 0
4 0 0 0 0 0 0
5 0 0 0 0 0 0
6 0 0 0 0 0 0
7 0 0 0 0 0 0
8 0 0 0 0 0 0
9 0 0 0 0 0 0
1 10 100 1000
LC
BR
AdoptOptAPOCentralized
![Page 23: Impact of Problem Centralization on Distributed Constraint Optimization Algorithms John P. Davin and Pragnesh Jay Modi Carnegie Mellon University School](https://reader035.vdocument.in/reader035/viewer/2022062715/56649d795503460f94a5be45/html5/thumbnails/23.jpg)
Conclusions
• Cycle-Based Runtime (CBR), a performance metric which accounts for both communication + computation in DCOP algorithms.
• Comparison of Adopt + OptAPO showing Adopt has a lower CBR at several communication latencies.
• Future Work– Compare DCOP algorithms on a distributed system.