Interaction Traces Mining for Efficient System Responses Generation

Abstract

Software service emulation is an emerging technique for creating realistic executable models of server-side behaviour. It is particularly useful in quality assurance and DevOps, replicating production-like conditions for large-scale enterprise software systems. Existing approaches can automatically build client-server and server-server interaction models of complex software systems directly from analysis of service interaction trace data. However, when these interaction traces become large, searching an entire trace library to generate a run-time responses can become very slow. In this paper we describe a new technique that utilises data mining, specifically clustering algorithms, to pre-process large amounts of recorded interaction trace data. With the obtained clusters we facilitate efficient yet well-formed runtime response generation in our Enterprise System emulation environment. We evaluate our approach using two common application-layer protocols: LDAP and SOAP. Our experimental results show that by utilising clustering techniques in the pre-processing step, the response generation time can be reduced by 99% on average compared with existing approaches.