A Systematic Approach towards Detecting, Identifying, and Recovering from Component Failures

The capability for recovery from failures is incorporated in almost every computer-based system that exists today. Yet, the design of fault tolerance mechanism for highly available, mission critical weapons systems is difficult, due to lack of coordinationbetween fault tolerance techniques. In some situations the fault tolerance strategies interact with or even interfere with each other, possibly leading to a cascade of failures, resulting in overall system failure. These issues must be addressed whenmodern weapons systems are created from commercial components.The crux of the issue is identifying the characteristics of a failure - the particular guarantee of service that has not been satisfied. The ability to specify this failure service policy, with defined, documented, and measurable parameters, will enable ananalytic approach to the problem. We propose to develop a dependability model for component interactions, along with an API that incorporates failure-relevant QoS parameters and supports implementations of the common failure detection strategies. We willuse this model in the context of the Navy Open Architecture to investigate the feasibility of managing real-time, mission-critical applications using commercial enterprise resource management systems within the NOA resource management system. The keybenefit of this work will be the reduction in complexity of developing failure management approaches for COTS-based weapons systems, with attendant reductions in time and cost of deployment. As the proposed work is leveraging COTS practice and standards,it will alos apply to commercial situations presenting complex failure management problems, such as digital control systems used in critical infrastructure management, and also large-scale distributed or grid computing activities.