This dissertation proposed and evaluated a new approach for generating realistic traffic
in networking experiments. Our construction relied on several components to form
a coherent solution to this problem:
The a-b-t model of source-level behavior, which provides a generic but detailed
way of describing source-level behavior that is applicable to any Internet application.
An efficient measurement method for accurately translating the packet header
trace of any arbitrary TCP connection into its a-b-t connection vector, even in the presence
of packet reordering and retransmission.
The source-level trace replay method for generating traffic in a closed-loop manner, which
provides a way of introducing fully reproducible synthetic traffic in
The ability to directly compare original traffic and its source-level replay, after
incorporating network parameters also derived from packet header analysis.
Such a comparison enables us to assess the realism of the synthetic traffic.
A method for resampling a-b-t connection vectors that supports both
the introduction of controlled variability in the generated traffic
and the predictable scaling of the offered load.
The rest of this chapter discusses these components8.1, highlighting some concrete contributions and
open questions, which could be the subject of future work. Our focus is on the larger scheme of
things, so we refer the reader to the summaries of each chapter for additional findings and
possible refinement of our methodology.