Aspects for improvement of performance in fault-tolerant software

We describe the use of aspect-oriented programming to improve performance of fault-tolerant (FT) servers built with middleware support. Its contribution is to shift method call logging from middleware to application level in primary-backup replication. The novelty consists in no burden being placed on application writers, except for a simple component description aiding automatic generation of aspect code. The approach is illustrated by describing how synchronization aspects are weaved in an application, and modifications of an FT-CORBA platform to avoid middleware level logging. Evaluation is performed using a telecom application enriched with aspects, running on top of the aspect-supporting platform. We compare overheads with earlier results from runs on the base-line platform. Experiments show a drop of around 40% of original overheads. This is due to methods starting execution before previous ones end, in contrast to ordering enforced at middleware level where methods are executed sequentially, not adapting to application knowledge.