Dynamic profit optimization for composite Web services with SLA's

In this paper we investigate sequential decision mechanisms for composite web services. After executing each sub-service within the sequential workflow, new decisions are based on the observed response times, taking into account the expected reward of subsequent services. As input for the decision-making process, typical Service Level Agreement parameters, such as costs, rewards, penalties and agreed response-time deadlines, are used. We propose a decision-making model for deciding, after receiving service at sub-service k, whether or not to proceed to sub-service k+1, based on the remaining ``response-time budget" and the corresponding expected rewards and penalties. For this model, we develop two algorithms for optimal stopping criteria: (1) a backward recursion algorithm, and (2) an algorithm based on a forward-looking decision. To evaluate the profit gain that can be obtained using these algorithms, we have benchmarked them against expected profit in the default case when no decision-making process is applied. Our results show that dramatic profit gains can be obtained by sequential decision making, that the backward recursion mechanism is optimal with respect to profit, and that the forward-looking decision approach makes less type-I errors in most cases.