Asymptotic approximations for TCP Compound

In this paper, we derive an expression for throughput of TCP Compound connections with random losses. We consider the case of a single TCP connection with random losses and negligible queuing, i.e., constant round trip time. For this case, an approximation of throughput has been derived earlier using a deterministic loss model. We provide a theoretical justification for this approximation. Further, we use results from the deterministic model to derive an approximation for TCP throughput under random losses. For this, we consider a sequence of appropriately scaled window size processes, indexed by p, the packet error rate. We show that as p goes to zero, this sequence converges to a limiting Markov chain. We use the stationary distribution of the limiting Markov chain to give an approximation for TCP Compound throughput for small p. We validate our model and approximations through ns2 simulations.