Accurate measurement of system call service times for trace-driven simulation of memory hierarchy designs

This paper presents a methodology for accurately measuring system call service times and process blocking times in a UNIX-like computing environment. Problems that may cause inaccurate measurements stem from the method used to charge processes for execution time, the difference in the time consumed during a system call and a time quantum or dispatch interval, and the instrumentation overhead included in the measurement program. The collected measurements for each system call are used to form a distribution from which process blocking times can be drawn or for which the mean can be used as the process blocking time.