Authors:
Leland Hovey
and
Mina Jung
Affiliation:
Syracuse University, United States
Keyword(s):
Multiagent, Evolution, Load-sharing, Scheduling.
Related
Ontology
Subjects/Areas/Topics:
Agents
;
Artificial Intelligence
;
Artificial Intelligence and Decision Support Systems
;
Computational Intelligence
;
Cooperation and Coordination
;
Distributed and Mobile Software Systems
;
e-Business
;
Enterprise Information Systems
;
Evolutionary Computing
;
Formal Methods
;
Grid Computing
;
Informatics in Control, Automation and Robotics
;
Intelligent Control Systems and Optimization
;
Internet Technology
;
Knowledge Engineering and Ontology Development
;
Knowledge-Based Systems
;
Multi-Agent Systems
;
Planning and Scheduling
;
Simulation and Modeling
;
Soft Computing
;
Software Engineering
;
State Space Search
;
Symbolic Systems
;
Technology Platforms
;
Web Information Systems and Technologies
Abstract:
Grid CPU load-sharing is a subclass of computational grid resource management. Its purpose is to improve
grid throughput – High Throughput Computing (HTC). The problem is load-sharing optimization state-space
can be quite large. This is because of two factors: the load-sharing optimization problem is NP-complete,
and a large volume of CPU-intensive loads can require thousands of Internet connected CPUs. Approximate
models can find near-optimal solutions to NP-complete problems. Multiagent coalition formation (MCF) is
a particular approximate game theoretic approach for these problems. We propose a new distributed MCF
(DMCF) model for Grid CPU load-sharing, DMCF grouping genetic algorithm (DMCF-GGA). This paper
presents the model in detail. It also compares this model with our existing model, DMCF-spatial. The comparison
consists of a discussion of the models’ similarities and differences, and a comprehensive empirical
evalution. The results of this study are the following: The optimi
zation search cost of DMCF-GGA is significantly
less than DMCF-spatial. DMCF-GGA has a linear relation between coalition size and search cost
(for high throughput). We have found preliminary lower and upper bound estimates for the effective coalition
size. We have also found the average job sizes required for the run time of DMCF-GGA to be 1% of the job
execution time.
(More)