On RTO timer implementation in TCP

Abstract

The retransmission timeout (RTO) timer used in TCP has long been standardized by the IETF in RFC2988, referred to as the TCP-RFC in this paper. Over the years, various deficiencies have been identified. In this paper, we focus on the implicit RTO offset problem, where the exact timeout limit of each packet is stretched by restarting the timer using the current timer value on the arrival of each acknowledgement (that acknowledges some new data). It would result in a slow timeout detection which unavoidably degrades the TCP throughput. In this paper, we first present a review of the TCP-RFC with special focus on the implicit RTO offset problem. Based on the insights obtained, we propose an enhanced RTO timer implementation, called E-RTO, for TCP. The implicit RTO offset is removed by mimicking the operation of a multi-timer implementation using a single timer. We then compare our E-RTO with TCP-RTO by simulations. We show that the faster timeout detection of our E-RTO leads to a throughput improvement of up to 3%. © 2010 IEEE.