Capturing TCP burstiness in light-weight simulations

Abstract

Burstiness in data traffic is emerging as a critical property that protocol analysis nowadays can no longer ignore. To preserve the kind of burstiness or scaling phenomena observed in aggregated TCP traffic, we develop a behavioral model that captures TCP’s window-based and closed-loop control. Through a novel modeling technique – exhaustive state exploration, we systematically examine each TCP states over a restricted range of connection length and packet loss. This restricted range covers the TCP behavior most common to web exchanges. When connections step outside this range (becoming too long or taking more than one loss) we preserve simulation accuracy by taking a abstraction fault and changing to a more detailed model. By counting packets with interarrival times falling into certain critical intervals – round trip time (RTT) or retransmission timeout (RTO), we are able to create finite state automaton (FSA) with states and transitions indicating rounds of back-to-back packet transmissions. We demonstrate that an FSA approximation of TCP can produce light weight simulation models of TCP suitable for background traffic, and that these models accurately reproduce multifractal scaling behavior in IP network traffic.