next up previous contents
Next: Beyond Comparing Connection Vectors Up: diss Previous: Summary   Contents


Reproducing Traffic

$\textstyle \parbox{0.2in}{\mbox{}}$$\textstyle \parbox{5.25in}{\begin{singlespace}\textit{Sometimes the appropriate response to reality is to go insane.}\end{singlespace}}$$\textstyle \parbox{0.2in}{\mbox{}}$
$\textstyle \parbox{5.25in}{\begin{flushright}--- \textsc{{Philip K. Dick (1928--1982)}}, Valis\end{flushright}}$

$\textstyle \parbox{0.2in}{\mbox{}}$$\textstyle \parbox{5.25in}{\begin{singlespace}\textit{Dissertations are not finished; they are abandoned.}\end{singlespace}}$$\textstyle \parbox{0.2in}{\mbox{}}$
$\textstyle \parbox{5.25in}{\begin{flushright}--- \textsc{{Frederick P. Brooks, Jr. (1931--)}}\end{flushright}}$

This chapter examines the statistical characteristics of source-level trace replay experiments, comparing them to those of their corresponding original traces. As discussed in Chapter 5, and illustrated in Figure 5.1, a packet header trace \(\mathcal{T}_h\) and its source-level trace replay can be compared at two levels. The first level is how well the set of connection vectors \(\mathcal{T}_c\) extracted from \(\mathcal{T}_h\) are preserved by the trace replay experiments. This means to collect a packet header trace \(\mathcal{T}_h^\prime \) from the replay and extract a new set of connection vectors \(\mathcal{T}_c^\prime \). Section 5.2 presented a comparison of \(\mathcal{T}_c\) and \(\mathcal{T}_c^\prime \) for three traces. It demonstrated that the characteristics of \(\mathcal{T}_h\) captured by \(\mathcal{T}_c\) are accurately reproduced by the traffic generation method and its implementation. The second level at which traces and their replays can be compared is to directly extract statistics from \(\mathcal{T}_h\) and \(\mathcal{T}_h^\prime \). If these statistics are reasonably close, we can say that the traffic generation method reproduces the original traffic using closed-loop traffic generation. This is the type of comparison discussed in this chapter. As we will show, source-level trace replay generally results in a good approximation of the statistical characteristics of the original traffic, which supports the use of the a-b-t model as a foundation for realistic traffic generation.



Subsections
next up previous contents
Next: Beyond Comparing Connection Vectors Up: diss Previous: Summary   Contents

Doctoral Dissertation: Generation and Validation of Empirically-Derived TCP Application Workloads
© 2006 Félix Hernández-Campos