Using Resampling Techniques to Compute Confidence Intervals for the Harmonic Mean of Rate-Based Performance Metrics

Rate-based metrics such as floating point operations per second, instructions per cycle and so forth are commonly used to measure computer performance. In addition to the average or mean performance of the metric, indicating the precision of the mean using confidence intervals helps to make informed decisions and comparisons with the data. In this paper, we discuss the determination of confidence intervals for the harmonic mean of rate-based metrics using two statistical resampling techniques - Jackknife and Bootstrap. We show using Monte Carlo simulations that resampling indeed works as expected, and can be used for generating confidence intervals for harmonic mean.