Distributed fault-tolerant methods for faulty hypercube and replicated database

Abstract

In this thesis, we investigate problems of distributed fault-tolerant method for a faulty hypercube and replicated database. The fault-tolerance can be provided by a combination of established hardware fault-tolerance and the fault-tolerant software techniques. We are mainly concerned with the fault-tolerant software techniques. We consider two distributed fault-tolerant methods in this thesis: One is a distributed shortest path search algorithm and an efficient fault-tolerant routing algorithm in the presence of faulty components in a hypercube. The hypercube computer system is a well known parallel processing system because of their structural regularity and powerful embedding properties. Our distributed shortest path search algorithm overcome the drawbacks of the depth-first and the best-first search methods, and always finds a shortest path in the presence of faults in hypercubes. Also, we show that the distributed shortest path search algorithm is efficiently applied to a fault-tolerant routing in faulty hypercubes with static faults. The porposed fault-tolerant routing algorithm attempts to route every message to its destination via a shortest path in the presence of an arbitrary number of faulty components. To reduce the amount of information at each node required for the shortest path routing, the unnecessary propagation of information on faulty components should be avoided. Therefore, when only backtracking is enforced at a node, all network information of that node are added to a backtracking message. Another is a replicated data concurrency control protocol for a replicated database. A replicated database is a distributed database in which multiple copies of some data items are stored redundantly at multiple sites. The replcation of a data is used for improving the availability and read performance of data. We present the replicated data concurrency control algorithm based on the dynamic voting scheme in the presence of site and network failures. In ...