COÊLHO, Á. V. S.; http://lattes.cnpq.br/0191048428072248; COÊLHO, Álvaro Vinícius de Souza.
Resumo:
The improvement and popularity of connectivity between computers have created an environment where resources used by organizations, in order to achieve the computing they want, don’t needed be in their local infrastructure. They can acquire computing from third parties, as a service. These services can be bought, but there is not only one way to obtain external computing services. For some applications, specially that ones that can be performed using a best effort strategy, an interesting alternative is to use reciprocation-based systems, where peers donate services to others by expecting to receive services from them in the future. It occurs that, in such systems, costs and utilities of services are valued differently by each peer, according to his interests and operative characteristics. Being rational agents, peers aim to find mechanisms to maximize the amount of advantage they receive from the system. Moreover, their interest in remaining in the system depends on whather they to achieve more utility than costs when exchanging services. Peers need to operate with profit. Considering that the system is based on reciprocation, the natural way to maximize profit is by selecting services that give the best possible relation between their costs and the utility achieved by reciprocity. Unfortunately, it is not feasible to implement an algorithm to find the best possible selection of services, due to the system complexity and indeterminism. Peers need to perform heuristic-based methods. In this work we show that the services selection has a strong impact on peers’ profitability. We show services selection heuristics that achieve good results, although this does not occur in all environment conditions, because they do not take care of the cost/utility management. We implemented hill-climbing approached heuristics in order to solve this problem. We showed that, in the service selection problem, it is not possible
to consistently evaluate heuristics and, due to this, we defined a methodology that allows
us to evaluate them by performing a comparison with a referential algorithm. Our results
show that it is possible for peers to have profitability in this environment, that is necessary
in order to keep them interested in remaining in the system, although the selections made by the proposed heuristics are still far from the best possible selection.