Improving TCP throughput over lossy links using protocol-level speculations

Abstract

The degradation of transport control protocol (TCP) throughput in networks with lossy links is mainly due to the coexistence of two types of losses, congestion losses and link corruption losses. This is very similar to processor performance degradation due to control hazards in CPU design. First, two types of loss events in networks with lossy links can be considered as two possibilities of a branching result (correct speculation vs. incorrect speculation) in a CPU. Secondly, both the problems result in performance degradations in their application environments, i.e., penalties (in clock cycles) in a processor, and throughput degradation (in bit per second) in TCP networks. This has motivated us to apply speculative techniques (e.g., speculating on the outcome of branch predictions), used to overcome control dependencies in a processor, to TCP algorithm design when lossy links are involved in TCP connections. The objective of this paper is to propose a protocol-level speculation based TCP modification to improve its throughput performance over lossy links. Simulation results show that our proposed algorithm significantly improves TCP throughput in a network with satellite links.