TCP Throughput Achieved by a Folded Clos Network Controlled by Different Flow Diffusion Algorithms
TCP Throughput Achieved by a Folded Clos Network Controlled by Different Flow Diffusion Algorithms
复制标题
DOI：
10.18178/ijiee.2020.10.1.714
复制
发表时间：
2020-03
期刊：
影响因子：
--
通讯作者：
S. Ohta
中科院分区：
文献类型：
--
作者：
S. Ohta
摘要-折叠Clos网络（FCN）是数据中心网络经常使用的拓扑结构。其性能取决于在输入/输出开关处执行的流扩散算法。过去已经提出了几种比传统随机路由更均匀地在链路之间扩散流的算法。期望这些算法防止TCP（传输控制协议）吞吐量由于业务拥塞而降低。然而，目前还不清楚这些流扩散算法是否是显着有效的避免TCP流的吞吐量下降相比，传统的随机路由。本文通过分组级计算机仿真研究了流量扩散算法对TCP吞吐量的有效性。结果表明，流扩散算法可以有效地减少TCP流的数量，其中的吞吐量下降。
Abstract —A folded Clos network (FCN) is the topology often used for data center networks. Its performance depends on the flow diffusion algorithm executed at the input/output switches. Several algorithms that diffuse flows more equally between links than conventional random routing have been proposed in the past. It is expected that those algorithms prevent the TCP (transmission control protocol) throughput from decreasing due to traffic congestion. However, it is unclear whether these flow diffusion algorithms are significantly effective for avoiding throughput degradation of a TCP flow compared with conventional random routing. This paper investigates the effectiveness of flow diffusion algorithms for TCP throughput through packet-level computer simulation. The results confirm that flow diffusion algorithms can efficiently reduce the number of TCP flows, the throughputs of which are degraded.