Adaptive Fault Tolerance in Multi-Computer Systems Using Data Compaction and Two-Level Voting

Abstract

Multi-computer systems offer good conditions for utilization of various fault tolerance principles. In order to increase reliability with little expense, this paper proposes an Adaptive Two-Level Voting scheme for multicomputer systems. This scheme enables a process-oriented dynamically determined fault tolerance by fault masking. The basic principle of the first voting level is a decentralized 2-out-of-N Signature Voting of exchanged process signatures generated by data compaction of N identical process copies on N computer nodes. It can tolerate N-2 faulty computer nodes. On the second voting level the results of the first voting level are compared by V-Version Programming using diverse software. For Two-Level Voting (V,N), the number of involved program versions V and the number of involved program copies N can be chosen. This paper describes four variants of Adaptive Two-Level Voting, a sequential, a parallel, a comparison variant and a very efficient dynamic variant. An estimation of efficiency of the suggested variants is also given.