Analysis of a Replication-Aware Transaction model by means of Stochastic Reward Networks

Replication-aware transaction model presents a technique in which failures in the scope of a transaction issued on a group of replicated objects are simply ignored. In this technique, the failed replicated objects are removed from their relevant object group and are re-created somewhere else if needed. Therefore, there is no need to roll the whole transaction back. Previous evaluation of the technique based on an implemented prototype on FT-CORBA and CORBA Transaction Service (OTS) has shown better transaction throughput compared to other methods. This paper re-evaluates this technique by modeling it using stochastic reward networks (SRNs). The performance measurements derived from running the SRN model differs from the one gained from the implemented prototype. We discuss these differences and analyze how the SRN model and the implemented prototype revealed interesting points which were not previously considered.