A distributed resource-trading model is proposed for users of a Grid computing architecture based on peer-to-peer technology, in which each node can either sell or buy computing services anywhere in the network by sending messages to its nearest neighbours. A mean field approximation suggests that it is possible for the market to settle to a stable price and its numerical predictions are compared against a multi-agent simulation, showing good agreement. The proposed system is then demonstrated to outperform a central server system in terms of scalability and average load per node. An adaptive system that can adjust to different parameters is also developed. Having shown that the distributed system is feasible, the question of how participants in the market can make use of the system is addressed. The market created by the trading agents is modelled as a Markov Chain, which is compared with the market state of the simulation. Finally, ways in which agents can trade in computing power and optimise their decisions are considered using Markov Decision Processes.
pubs.doc.ic.ac.uk: built & maintained by Ashok Argent-Katwala.