A performance analysis of a buddy system for fault tolerance

A model for fault tolerant computing in a distributed computing system is presented and analyzed. Each time a job is submitted, two copies of it are stored: one at its original node, where it will normally be executed, and the other at a second node, called the buddy node. If the original node fails, the copy at the buddy node will be executed, providing fault tolerance. By means of an iterative procedure, the average queue length and the average response time may be calculated, with some simplifying assumptions. Comparison with simulation results shows excellent agreement. Numerical results are presented to show the effects of varying the parameters on the performance of the system.