Results

For evaluating the overall methodology and the policies outlined in Sections 3.3 and 4, we define the peak rate $\lambda ^*$ to be the test load that causes: (a) the mean server response time to be in the $[36,44]$ ms region; or (b) the 95-percentile request response time to exceed $2000$ ms to complete. We derive the $[36,44]$ region by choosing mean server response time threshold at the peak rate $R_{sat}$ to be $40$ ms and the width factor $s = 10\%$ in Table 2. For all results except where we note explicitly, we aim for a $\lambda ^*$ to be accurate within $10$% of its true value with $95$% confidence.