Technical Report: Service variant discovery using a likelihood-free Bayesian search method

Despite the proliferation of service interfaces available on the Internet, the integration of services from larger software systems remains hampered by large and overloaded operational signatures carrying hundreds of parameters, with multiple levels of nesting, and several artifacts, typically corresponding to inter-related business entities. We undertake, in this paper, to advance recent contributions to code-based service analysis, in support of interface refactoring, composition and run-time interface introspection. Specifically, we address the notorious problem of service variant derivation from single operations where subsets of parameters correspond to multiple nested business entity sub-types. Given the complexity of deriving sub-types using brute-force invocations over large parameter search spaces, we propose a Monte Carlo sampling method, based on likelihood-free Bayesian sampling, to identify higher probabilistic spaces for testing calls for prospective sub-types. Results presented herein demonstrate a method with significant success rates in massive search spaces, scaling up to successful derivations of variants in a simulated FedEx Shipment interface.