Search for an object with a recognition error probability in a Desktop Grid environment

In scientific computing, blind search problems are quite common. Objects from a finite set are independently studied one after another until a unique target is found. If the recognition may fail, so that the target may be missed, objects must be re-studied, according to some strategy. In Desktop Grid computing (Enterprise Desktop Grid, Volunteer computing) replication is commonly used to increase the reliability of a computing system. Doing the work twice or more times is a sacrifice of performance for reliability, and the optimal trade-off is not trivial. In this paper, we construct a mathematical model of this problem, from which we obtain the optimal trade-off policy, which turns out to be “no replication”: objects are re-studied not earlier than all of them have been studied once and no target has been obtained. We present an explicit formula for the expected number of examinations needed to find the target. The results of numerical experiments illustrate the result and compare the optimal strategy with other ones. Finally, we discuss the consequences of the results to Desktop Grid computing.