An Extended Model for TCP Loss Recovery Latency with Random Packet Losses

It has been a very important issue to evaluate the performance of transmission control protocol (TCP), and the importance is still growing up because TCP will be deployed more widely in future wireless as well as wireline networks. It is also the reason why there have been a lot of efforts to analyze TCP performance more accurately. Most of these works are focusing on overall TCP end-to-end throughput that is defined as the number of bytes transmitted for a given time period. Even though each TCP’s fast recovery strategy should be considered in computation of the exact time period, it has not been considered sufficiently in the existing models. That is, for more detailed performance analysis of a TCP implementation, the fast recovery latency during which lost packets are retransmitted should be considered with its relevant strategy. In this paper, we extend the existing models in order to capture TCP’s loss recovery behaviors in detail. On the basis of the model, the loss recovery latency of three TCP implementations can be derived with considering the number of retransmitted packets. In particular, the proposed model differentiates the loss recovery performance of TCP using selective acknowledgement (SACK) option from TCP NewReno. We also verify that the proposed model reflects the precise latency of each TCP’s loss recovery by simulations. key words: transmission control protocol (TCP), congestion control, loss recovery, performance modeling and validation, fast recovery latency