TCP Wave estimation of the optimal operating point using ACK trains

Abstract

TCP Wave changes the typical TCP transmission paradigm by replacing the ACK-clocked sliding window with self-scheduled bursts. In response to bursts, unmodified TCP receivers generate ACK trains, which carry useful information about the end-to-end link characteristics. TCP Wave inspects ACK-train flow to measure the following parameters: (i) ACK train spread (namely ACK train dispersion) and (ii) RTT variations. The former is expected to provide the overall “service capacity”, meant as the maximum capacity allowed over the end-to-end path, while the latter is considered as a congestion indicator. The joint use of such measurements allows TCP Wave to fine-tune the transmission rate to accurately match current network resource availability in the bottleneck link. This paper analysis TCP Wave ACK train-based measurements on a broad set of simulated links compliant to characteristics of today's real networks. To this scope, a testbed with TCP Wave implementation on Linux OS is used to perform tests varying both bottleneck capacity and physical latency.