Task Redistribution in Faulty Networks using Evolutionary Strategies

Fault tolerant distributed systems must be able to continue operation in the presence of hardware faults (albeit with degraded performance). If one or more processors fail, the tasks assigned to those processors must be reassigned to other operational processors. The recovery mechanism thus reduces to a task allocation problem in a faulty network. This problem is known to be NP -complete. In this paper we redistribute tasks in faulty networks using evolutionary strategies. We show that this technique is able to quickly determine a good reallocation of tasks with little degradation of performance. 1 Introduction Hardware faults may reduce the availability or usability of a distributed computing system. Systems which continue to operate (albeit, with a degraded performance) in the presence of hardware failures are referred to as fault tolerant systems. The goal of any fault tolerant system is to recover from the fault as quickly as possible with a minimal performance degradation. Dupli..