Experimental performability evaluation of middleware for large-scale distributed systems

Abstract

We present a tool for experimental evaluation of distributed software that enables performability tests to be incorporated in the agile development of complex middleware systems. The proposed approach combines simulation with the profiling of key components and is achieved by an extension to a standard simulation kernel and reflection, thus leveraging existing simulation models and easing the integration of existing components. The evaluation of database replication middleware in large-scale systems is used to illustrate the approach.