An algorithm to detect TCP spurious timeouts and its application to operational UMTS/GPRS networks

摘要

This paper proposes an algorithm to identify TCP spurious retransmission timeouts by post processing of packet traces monitored in operational networks. The operational principles of the algorithm and the assumptions behind its design are explained in detail as well as the situations in which the algorithm is prone to inaccuracies. By extensive measurements in a lab testbed using realistic round trip time characteristics as observed in operational wireless networks and FTP-like as well as Web-like traffic generators, it is shown that the algorithm is accurate in detection of spurious retransmission timeouts. Subsequently, the algorithm is applied to real traffic traces captured at several interfaces of an operational UMTS and GPRS network to analyze the frequency of spurious retransmission timeouts as well as the spurious timeout probability dependent on the load situation in the network and the flow size. This investigation, to our best knowledge the first on large scale TCP traffic traces monitored in an operational UMTS network, shows that spurious timeouts are infrequent events in the considered UMTS as well as the GPRS network. Among other findings, it is additionally shown that the ratio between spurious timeouts and other congestion recovery events experienced by TCP flows is low, indicating a negligible impact of spurious timeouts on TCP performance.