MUTEAM : a multimicroprocessor architecture with decentralized fault treatment

MUTEAM is aimed at experimenting in the field of decentralized control policies for error treatment and resource management. This paper contains a brief discussion of the design philosophy, followed by a description of the hardware, operating system kernel and fault-tolerance mechanisms. The fault-treatment policies are then described. The hardware architecture is composed of a set of clusters: each cluster is a tightly-coupled multiprocessor. Within a cluster, specialized hardware is provided for memory protection and management, and for interprocessor communication. The kernel implements the run-time support for a csp-like, message-based concurrent programming language. The implemented prototype includes an integrated set of fault-tolerance mechanisms, inserted in the system programming language and run-time support, and in the physical architecture. These are used to develop a fault-treatment policy, consisting of separate phases of diagnosis, reconfiguration and recovery. All of these are executed by sets of parallel processes, without a supervisor entity. Diagnosis is provided, in particular by a set of dedicated processes, one of which runs in each computer element, while reconfiguration and recovery are obtained through instrumentation of the application code. 100 references.