Cloud computing is placing increasingly stringent demands on datacenter networks. Applications like MapReduce and Hadoop demand high bisection bandwidth to support their all-to-all shuffle communication phases. Conversely. Web services often rely on deep chains of relatively lightweight RPCs. While HPC vendors market niche hardware solutions, current approaches to providing high-bandwidth and low-latency communication in the datacenter exhibit significant inefficiencies on commodity Ethernet hardware. We propose addressing these challenges by leveraging the tightly coupled nature of the datacenter environment to apply time-division multiple access (TDMA). We design and implement a TDMA MAC layer for commodity Ethernet hardware that allows end hosts to dispense with TCP's re-liability and congestion control. We evaluate the practicality of our approach and find that TDMA slots as short as 100s of microseconds are possible. We show that partitioning link bandwidth and switch buffer space to flows in a TDMA fashion can result in higher bandwidth for MapReduce shuffle workloads, lower latency for RPC workloads in the presence of background traffic, and more efficient operation in highly dynamic and hybrid optical/electrical networks.
展开▼
