Finding the Peak Rate

In the inner loop of Algorithm 1, the automated controller searches for the peak rate $\lambda ^*$ for some workload and configuration given by a selected sample of factor values in $\langle F_1, \ldots, F_n \rangle$. To find the peak rate it subjects the server to a sequence of test loads $\lambda = [\lambda_1, \ldots, \lambda_l]$. The sequence of test loads should converge on an estimate of the peak rate $\lambda ^*$ that meets the target accuracy and confidence.

We emphasize that this step is itself a common benchmarking task to determine a standard rating for a server configuration in industry (e.g., SPECsfs [6]).