A market-based approach to the dynamic reconfiguration problem of service-based systems

Modern complex distributed systems are built as service-based systems (SBSs), composed of Web services, which are discovered and bound (most of the time) during runtime to handle the dynamic nature of Web service offerings. As far as vie know, all reconfiguration approaches assign to SBSs' owners the responsibility to identify, select and compose services that satisfy their reconfiguration needs. In practice, SBSs are not omniscient regarding potential service providers' partners or offer trade-offs; thus, in most (or all) cases, service compositions to reconfigure SBSs are suboptimal. This article describes the market-based service reconfiguration (MBSR) approach, where service providers and requesters are software agents and reconfigurations naturally arise from the interaction of them. We implement MACOCO++, a multi-agent reference implementation to configure and reconfigure SBSs; the studied SBSs yield systematically better reconfigurations, and the prototype scales up well regarding both, complexity and number of concurrent service requests. Our results show that using an agent-based market metaphor yields better, fairer service agreements for requesters without needing a central (and possibly unfeasible) component with global knowledge.