Smarter Heuristics for Business-Driven Services Selection in Multi-services P2P Grids

In peer-to-peer (P2P) grids, peers act both as providers and consumers of the services offered in the system. In these systems, fair and efficient match of service request and service provision can be attained by using simple reciprocation-based mechanisms. Under contention, a peer p gives priority to serve the requests of the peers that have the largest difference between the amount of service provided to p and the amount of service consumed from p in the past, i.e. the peers to whom p owes most. Taking a business-driven approach, peers have a cost for the provision of services, while gain utility for the consumption of services, and the peers' profit is given by the difference between the overall utility attained and the overall cost incurred. It has been shown that if a peer may offer multiple services, then the profit that it can extract from the P2P grid is highly affected by the services that it chooses to offer. Thus, the services selection problem can be formulated as an optimisation problem that seeks to maximise the peers' profit. Unfortunately, due to the many uncertainties that characterise this highly dynamic system, it is not feasible for a peer to deterministically decide what is the optimal selection of services that it should make. Thus, services selection algorithms must be based on heuristics. Some heuristics have been proposed, but although they perform well in a number of scenarios, there are many others in which they fall short. This is basically due to their inability to identify situations in which it would be better not to use their resources to provide services that will not yield any profit in the future. In this paper we use a hill-climbing approach to design smarter heuristics that can provide more profitable selections, when compared to previously proposed ones. Our simulation results show that, in the scenarios evaluated, the new heuristics proposed never perform worse than the best of the previously proposed ones, and can outperform the latter in as much as 281% in the most favourable settings when resources are plentiful.