Fault tolerant computing in computational field model

Abstract

In large scale distributed systems, fault tolerant computing is important because each module may not be always reliable. Fault tolerant computing is not essentially needed to solve a problem. However, it is useful to execute a computing correctly. In this paper, we propose the usage of computational field model (CFM) as a framework to reuse such computing. Computational field is shared virtual space which abstracts distributed systems. It is possible to construct portable applications by applying algorithms to CFM. At first, we employ Triple Module Redundancy (TMR) as basic technique for fault tolerant computing in order to support real-time applications. Next, we assume the locality of fault occasion. For an example, physical crash causes faults locally. In such a case, each module should be distributed to increase system reliability. However, when they are distributed, system performance may be decreased because communication cost is increased. Thus, fault tolerance is related to system performance. In our approach, it is possible to accomplish both resource allocation and fault tolerant computing at the same time.