Dimensions for Reliability Contracts in Distributed Object Systems

Designing and managing distributed systems with predictable reliability and availability is generally difficult. Whenever components are specified, used, and managed it is often unclear what reliability requirements different components are expected to satisfy. The problem of specifying and satisfying reliability requirements needs to be addressed for many different situations and contexts. We need languages and tools that allow design time and run time specification of reliability requirements and offerings. We need quality of service (QoS) contracts that can be negotiated and monitored dynamically. From a management view, we need the ability to define reliability contracts between system components so that they can be effectively managed and charged for in heterogeneous and federated systems. Common to all these situations is the need for a vocabulary for the specification of reliability characteristics. Such a vocabulary should be complete in the sense that it captures the important aspects of service reliability without getting too complicated. In addition, it should be independent of any implementation techniques and mechanisms. This paper proposes such a vocabulary as a set of dimensions that can be used to characterize the reliability of distributed services. The dimensions are focused on distributed object systems but can easily be abstracted to cover other architectures for distributed systems.