Performability Modeling for Multi-layered Service Systems

Distributed software is usually structured in layers, with some kind of operational control or user interface tasks as the topmost layer, making requests to various layers of servers. Client Server systems and ODP (Open Distributed Processing) systems such as DCE, ANSA and CORBA are structured this way, as are distributed transaction processing systems. Performance and dependability are the two strong motivators for introducing distributed processing. A special form of layered performance model called Layered Queueing or Rendezvous Nets has been developed to model such systems, to provide a model formulation which is close to the software architecture (e.g. [1, 2, 3]). This note describes how dependability analysis could be added to the layered model, in a way which would make the dependability-related parameters particularly easy to specify. The result would be a straightforward and powerful performability analysis for distributed software systems. Performability concepts and tools are described in many papers, for example a recent survey by Haverkort and Niemegeers [4].