The TCP Outcast Problem: Exposing Unfairness in Data Center Networks

In this paper, we observe that bandwidth sharing via TCP in commodity data center networks organized in multi-rooted tree topologies can lead to severe unfairness, which we term as the TCP Outcast problem, under many common traffic patterns. When many flows and a few flows arrive at two ports of a switch destined to one common output port, the small set of flows lose out on their throughput share significantly (almost by an order of magnitude sometimes). The Outcast problem occurs mainly in taildrop queues that commodity switches use. Using careful analysis, we discover that taildrop queues exhibit a phenomenon known as port blackout, where a series of packets from one port are dropped. Port blackout affects the fewer flows more significantly, as they lose more consecutive packets leading to TCP timeouts. In this paper, we show the existence of this TCP Outcast problem using a data center network testbed using real hardware under different scenarios. We then evaluate different solutions such as RED, SFQ, TCP pacing, and a new solution called equal-length routing to mitigate the Outcast problem.