Batching for Smart Home: Leveraging delay-insensitive workload in cloud storage

Abstract

We study the problem of managing high intra-datacenter traffic in a chunk-based public cloud storage service. The high traffic volume is introduced by aggregating very large files from multiple chunk servers in a single edge server. We measure a commercial cloud storage service system, and observe that peak traffic volume overwhelms the network interface cards (NICs) significantly. In the scenario of delivering content based on Smart Home network, it can be expected the file downloading service could be delay-insensitive. Thus, we propose “Batching Smooth intra-datacenter Traffic” (BST) scheme to reduce the peak load to a specified upper bound by batching and delaying users' requests. We resort to a mathematical model to understand the necessity of batching strategy. To evaluate BST's effects, we implement trace-driven simulations with different scheduling policies. In the commercial cloud storage service system, we show that BST is capable of keeping the upper bound to approximately 75% of the original peak traffic by trading off an average delay of 8 minutes.