Orchestrating Unreliable Services: Strategic and Probabilistic Approaches to Reliability

Abstract

The robustness of orchestrations of unreliable web-services to failure can be analysed using Angel Daemon ($\mathcal A/\mathcal D$) games. A measure of the reliability of an orchestration O can be determined by characterising a user’s perception of underlying services and, additionally, making assumptions about the number of services that will fail during an evaluation of O (here such characterisations are called uncertainty profiles). The Nash equilibria of games associated with uncertainty profiles provide a posteriori information about the probability of orchestration success. In this approach probabilities are second class objects since they are derived from calculated Nash equilibria. Alternatively, probabilistic characterisations of failing services can be given from the outset (probabilistic profiles). Uncertainty profiles and probabilistic profiles provide complementary techniques for investigating the robustness of web-based orchestrations. In this paper the relationship between the two approaches is investigated. A means of adding probabilistic information to an uncertainty profile is proposed – hybrid profiles give rise to bayesian variations of $\mathcal A/\mathcal D$ games. The main result of the paper is to align $\mathcal A/\mathcal D$ games with bayesian $\mathcal A/\mathcal D$ games. When the probabilistic information retrieved by the angel is consistent with a Nash equilibrium, the corresponding bayesian Nash equilibrium mimicks the original Nash equilibrium.

J. Gabarro and M. Serna are partially supported by TIN-2007-66523 (FORMALISM), and SGR 2009-2015 (ALBCOM). Alan Stewart is partially supported by EPSRC project EP/I03405X/1 (ECHO).