On the properties of an adaptive TCP Minimum RTO

The Internet Engineering Task Force (IETF) specified in RFC 2988 the Minimum TCP–RTO and recommended that the TCP Retransmission Timeout (RTO) should not be smaller than 1 s. According to RFC 2988, there are two main limitations that call for a lower bound to protect TCP from spurious timeouts: (i) the OS clock granularity (500ms for most OSs at the time of RFC 2988 publication) and (ii) the Delayed Acknowledgments (usually set to 200ms).We evaluate the correctness of the suggested policy and investigate the impact of the conservative 1-s Minimum TCP–RTO under modern networking conditions. We define a Cost Function to capture the impact of the extra waiting time, due to the conservative IETF specification. Our experimental analysis reveals that the OS clock granularity should not be a matter of concern for modern OSs; we carefully design a mechanism to deal with Delayed ACKs. We call the proposed mechanism Adaptive MINRTO (AdMINRTO) to reflect its operational properties. In particular, the mechanism identifies the packets whose ACKs are (possibly) going to be delayed and applies extended Minimum RTO to these packets only, in order to avoid spurious timeout expirations; otherwise, the Minimum RTO is adjusted to smaller values in order to avoid extensive idle periods.We show, through simulations, that the Adaptive MINRTO improves TCP performance significantly, especially in case of next generation’s high-speed, wireless-access networks. The impact of the proposed mechanism, performance-wise, depends on several network conditions. For example, the impact of the Adaptive MINRTO increases (i) with the Packet Error Rate (PER), (ii) when the path Round-Trip Time is short and (iii) for fast transmission links (i.e., high-speeds).