The results in this chapter demonstrated that source-level trace replay can closely approximate the characteristics of real traffic traces. We have also shown that full source-level replays are closer or far closer to original traces than collapsed-epochs replays for several metrics. In particular, the largest difference is observed for the time series of packet throughput, the body of the packet throughput marginal and the time series of active connections. Byte throughput is similar for full and collapsed-epochs replays. The latter exhibits somewhat more bursty time series, but the bodies of the marginals do not change significantly.
The rest of the metrics cannot be clearly interpreted, since losses have a much more significant impact on them than the use of full or collapsed-epochs replays. Lossy full replays are clearly better than lossy collapsed-epochs replays in terms of wavelet spectra, estimated Hurst parameters and tails of the marginals for some traces, but this is not consistent for the five traces. Our analysis clearly demonstrated the need to carefully consider the impact of losses on evaluating the quality of synthetic traffic. Without our direct comparison of lossless and lossy replays, the results for certain metrics could have mislead our conclusions regarding source-level modeling. In contrast, other metrics are less affected by the loss model. This is the case for the time series of packet throughput, the body of the packet throughput marginal and the time series of active connections, where full replays are clearly better approximations than collapsed-epochs replays.
Doctoral Dissertation: Generation and Validation of Empirically-Derived TCP Application Workloads
© 2006 Félix Hernández-Campos