 |
In Figure 10, we compare the performance from the two active measurement based techniques (i.e., SlidingWindow and FrequencyCounts) with the passive measurement approach. Since our earlier results showed that history does not help in improving performance, henceforth we present results in which no history is employed. We compare the performance of the three measurement schemes for a common sampling interval of 120s across the five client workloads.
Note that the two active measurement schemes offer comparable performance. Unfortunately, the workloads we selected do not bring out other underlying trade-offs of these schemes (discussed earlier in Section 3.1.2). A detailed comparison of these active measurement schemes is future work.
Figure 10 also shows that the active measurement-based schemes offer slightly better performance than the passive measurement scheme: about 8-10% for the light workloads and 2-3% for the heavier workloads. This is expected, since the passive scheme uses existing transfers to obtain samples of performance across the, potentially sub-optimal, ISP links.
 |