Performance Analysis of Instrumentation System Management Policies

Run-time trace data helps to debug and analyze the parallel programs. Obtaining and managing this data during run-time is the responsibility of an instrumentation system that incurs overhead. In its worst case, this overhead can result in severe perturbation of the behavior of the actual program. This paper presents a queuing model for an instrumentation system. The purpose is to provide a rigorous mathematical tool that could allow the analysis of the overhead of program behavior due to the instrumentation system management policies. We summarize the effects of two management policies: FOF and FAOF policies.