SA-TCP: A Novel Approach to Mitigate TCP Incast in Data Center Networks

Abstract

In high bandwidth and low latency data center networks, when multiple data senders simultaneously communicate with a single receiver, namely in many-to-one communication pattern, it is frequently occurred that the burst data overload the receiver's switch buffers, which leads to TCP throughput collapse. This phenomenon is termed to TCP Incast problem, which has become a new hot research topic in recent years. In this paper, we propose a novel TCP congestion control algorithm named Stochastic Adjustment TCP (SA-TCP) which can stochastically adjust the congestion control window to avoid the synchronization among many parallel flows so as to reduce the occurrences of congestion and mitigate the TCP Incast problem. The simulation results based on NS2 verify that SA-TCP has higher goodput and can support a larger number of servers transferring concurrently than TCP. Moreover, SA-TCP has good convergence and fairness performance. It can mitigate the TCP Incast problem in data center networks effectively.