Loss Distinguishing Accuracy in TCP Veno and its Performance Influence.
Ke Zhang, Cheng Fu, Zixuan Zou · 2006
In this paper, we aim to study TCP Veno's loss distinguishing accuracy in different network conditions and its performance influence. We first propose a heuristic formula. Based on the simple math analysis and experiment results, we show the sensitivity of Veno's loss distinguishing accuracy on network parameters. Thereafter, we demonstrate two cases to study its performance influence. One case is accuracy extremely high, and another is accuracy extremely low. We observe that, even if the loss distinguishing accuracy is low, Veno still performs well without negative effect. Detailed analysis show the reason is, timeout may play the main role in TCP performance and Veno inherits from Reno the action for timeout without modification. performance influence. We first propose a heuristic formula. Based on the simple math analysis and experiment results, we show the sensitivity of Veno's loss distinguishing accuracy on network parameters. Thereafter, we study the performance influence by comparing NewVeno and NewReno in two cases. NewReno is an enhancement of TCP Reno by Partial ACK to avoid multiple window reduction in the recovery phase (9). NewReno is considered because it is supported by most of the TCP implementations in Internet. For a fair comparison, we also add the Partial ACK into Veno and call it NewVeno. This modification doesn't affect the loss distinguishing accuracy. The first case is the wireless scenario, in which the loss distinguishing accuracy is extremely high and TO is negligi- ble. NewVeno achieves significant performance improvement over NewVeno duo to salient accuracy. The second case is NewVeno and NewReno co-existing in the wired network, in which the loss distinguishing accuracy is extremely low but NewVeno harmoniously co-exists with NewReno without any negative impact. The detailed analysis shows that TO in place of TD plays the key role in this case. NewVeno is compatible with NewReno because Veno inherits from Reno the action for TO without modification. The rest of this paper is organized as follows. In section II, we briefly introduce the TCP Veno mechanism and our simu- lation topology. Section III is the analysis of TCP Veno loss distinguishing accuracy under different network parameters. In section IV, we use two cases to study the influence of loss distinguishing accuracy. Section V is the conclusion.