Practical Issues with Replay of High Throughput Traffic

摘要

Traffic replay recreates conditions and states found in traffic trace repeatedly within an emulation or a testbed. When throughput in the original traffic trace is high, it becomes impossible to replay it using a single transmitter-sink pair of machines. The only solution is to distribute the replay process among several replay sources, which in turn requires the traffic trace to be split in several parts. This paper proposes a solution for such distributed replay by applying the notion of traffic type and traffic entity, both of which are high-level abstractions of individual packets in traffic. The paper also studies how evenly the replay burden can be distributed using an arbitrary number of artificial traffic sources.