DiFS: Distributed Flow Scheduling for Data Center Networks

摘要

Data center networks leverage multiple parallel paths connecting end hostpairs to offer high bisection bandwidth for cluster computing applications.However, state of the art distributed multi-pathing protocols such as EqualCost Multipath (ECMP) use static flow-to-link assignment, which isload-oblivious. They may cause bandwidth loss due to \emph{flow collisions} ona same link. Recently proposed centralized scheduling algorithm or host-basedmulti-pathing may suffer from scalability problems. In this paper, we present Distributed Flow Scheduling (DiFS) for data centernetworks, which is a switch-only distributed solution. DiFS allows switchescooperate to avoid over-utilized links and find available paths withoutcentralized control. DiFS is scalable and can react quickly to dynamic traffic,because it is independently executed on switches and requires nosynchronization. Extensive experiments show that the aggregate bisectionbandwidth of DiFS using various traffic patterns is much better than that ofECMP, and is similar to or higher than that of a recent proposed centralizedscheduling algorithm.