利用多路径传输协议,多宿主主机可以通过多条路径并行传输数据,从而有效提高系统的吞吐率和鲁棒性.但是由于不同路径在带宽、延迟和丢包率等方面存在差异,接收端必须缓存大量乱序到达的分组.数学分析表明,减少接收端的缓存开销有两条途径:一是最小化每条路径的发送队列中积压分组的数量,二是降低分组发送速率.由前者,提出依据每条路径的空闲发送窗口大小进行分组调度的算法SOD(Scheduling On Demand);由后者,提出利用窗口通告机制限制分组发送速率的流控方法.模拟实验结果表明:与现有算法相比,SOD的缓存开销最小;在接收端进行流控限制的情况下,SOD的吞吐率最大,并且在不同实验场景中性能表现稳定.%With the aid of multipath transport protocols, a multi-homed host can transfer data through multiple paths in parallel to improve goodput and robustness. However, the receiver has to deal with a large quantity of out-of-order packets due to the discrepancy of paths in terms of bandwidth, delay and packet losses. Theoretical analysis suggests that there are two approaches to reducing the memory overhead of caching out-of-order packets. One is to minimize the quantity of packets backlogged in outgoing queues of senders, and another is to decrease the packet sending rate. From the former, the study proposes a packet scheduling algorithm, named SOD (Scheduling On Demand), which assigns packets to each path according to the free window size. From the latter, a simple flow control method is proposed, which leverages the window feedback advertisement mechanism to limit the packet sending rate. Experimental results show that compared with existing algorithms, SOD suffers from the lowest memory overhead and obtains the highest goodput when receivers enable flow control. Additionally, SOD works steadily in the cases of diverse simulation scenarios.
展开▼
