Theoretical and experimental results of processing N tasks in the presence of processor failures

This paper presents a comparison between experimental and theoretical results of distributed applications that perform N independent and idempotent tasks using P fault-prone processors in parallel. The two are compared to see how significant are certain overhead factors that are not included with the analytical model, i.e. communication costs and overhead of other activities of the system. It is shown that the difference between the experimental and theoretical results is a reasonable measure of actual overhead of the prototype, but there are some interesting discrepancies.