摘要： 针对TFRC协议与TCP协议超时重传参数RTO计算方法不同而导致的两协议竞争时吞吐量差异较大的问题，提出一种基于概率偏差的TFRC-RTO算法。该算法以实际丢包事件概率与最优丢包事件概率之间的偏差为依据，对实时TFRC-RTO的值进行放大或缩小，使其更接近当前的TCP-RTO。仿真实验表明，该算法对TFRC协议具有明显的改进作用，与原算法相比，在短延迟网络中吞吐量提高了50%以上，报文丢失率减少了一半。

关键词: TFRC协议, 超时重传, 概率偏差, 拥塞控制

Abstract: To the question of different calculating methods between TFRC-RTO and TCP-RTO result in a great difference in throughput when they are competing, this paper presents an algorithm based on probability bias for TFRC-RTO. The algorithm zooms in or out the TFRC-RTO through subtracting the probability of actual packet loss events from the probability of optimal packet loss events to make it much closer to TCP-RTO. Simulation results indicate that, compared with the original algorithm, the new RTO algorithm enhances above 50% in throughput and the package loss number is reduced by half with short-RTTs.

Key words: TCP-Friendly Rate Control(TFRC) protocol, retransmission time-out, probability bias, congestion control

中图分类号: 

• TP393