Hierarchical error detection in a software implemented fault tolerance (SIFT) environment

A key problem besetting distributed applications is how to provide reliability guarantees to them, running on off-the-shelf hardware and software components. Chameleon is a Software Implemented Fault Tolerance (SIFT) middleware capable of providing adaptive fault tolerance in a COTS (components-off-the-shelf) environment with the capability to adapt to changing runtime requirements as well as changing application requirements. The thesis presents the architecture and implementation of a hierarchy of error detection techniques, which can be applied in a distributed SIFT environment. The error detection framework is implemented and demonstrated on the Chameleon testbed, though the principles are of general applicability in a message-passing-based distributed system. The thesis shows how the detection mechanisms are applicable to the components of the SIFT layer and how to extend them to applications executing on such a layer. A flexible mechanism for combining the different levels in the hierarchy to ensure the environment's adaptivity is presented. Next, the thesis presents some novel detection techniques targeted to different kinds of faults—control faults, data faults, message faults, etc. Particularly, a new approach called software signatures, which is used for validating the integrity of software components in Chameleon, is presented. The thesis also presents results from fault injection based assessment of the detection protocols and their performance measures.