Improving Interaction with Services via Probabilistic Piggybacking

Modern service oriented applications increasingly include publicly released services that impose novel and compelling requirements in terms of scalability and support to clients with limited capabilities such as mobile applications. To meet these requirements, service oriented applications require a careful optimisation of their provisioning mechanisms. In this paper we investigate a novel technique that optimises the interactions between providers and clients called probabilistic piggybacking. In our approach we automatically infer a probabilistic model that captures the behaviour of clients and predicts the future service requests. The provider exploits this information by piggybacking each message toward clients with the response of the predicted next request, minimizing both the amount of exchanged messages and the client latency. The paper focuses on REST services and illustrates the technique with a case study based on a publicly available service currently in use