Building and Evaluating a Fault-Tolerant CORBA Infrastructure

In this paper we explore the trade-offs involved in making one major middleware fault-tolerant. Our work builds on the FT-CORBA specification (April 2000), which is not in widespread use due to lack of quantified knowledge about design rules and trade-offs. Recent results show preliminary overhead and trade-off analysis for a not entirely FT-CORBA-compliant system for embedded applications (Eternal). In distinction from Eternal, we have extended an existing open source ORB and combine it with a collection of service objects and portable request interceptors following the FT-CORBA standard. The paper reports on extensive studies relating the timing aspects to different parameters such as the replication style, the number of replicas, and the checkpointing frequency. The experiments were conducted using a realistic telecom application.