Engineering elastic traffic in TCP-based networks: Processor sharing and effective service time

Abstract

Today, a wide range of networks exist that provide users with data services using TCP. For performance engineering of such networks there is a need for simple but accurate models to predict the performance under anticipated load conditions. This is a challenging task, because the combined packet-level dynamics of the network-protocol stack are highly complicated. First, we explain how the packet-level dynamics and protocol overhead of the multiple communication layers can be successfully captured by a single parameter, called the effective service time (EST), and give a full parameterization that explicitly expresses the EST in terms of the network parameters. Second, we show how the transfer-time performance can be modeled by a classical M/G/1 Processor Sharing (PS) model, but where the service time is replaced by the EST. Finally, extensive test-lab experimentation shows that the model leads to highly accurate predictions over a wide range of parameter combinations, including light- and heavy-tailed file-size distributions and light-and heavy-load scenarios. The simplicity and accuracy of the model make the results of high practical relevance and useful for performance engineering purposes.