Computer Science Thesis Proposal

Tuesday, November 17, 2015 - 2:00pm


Traffic 21 Classroom 6501 Gates & Hillman Centers


TIMMY ZHU, Ph.D. Student

Meeting tail latency Service Level Objectives (SLOs) in shared networked storage systems is an important and challenging problem in datacenters. Our work is motivated by three trends: First, companies like Google and Amazon are increasingly interested in long tails at the 99th and 99.9th percentile latencies. As technology improves, users are more accustomed to low latency and start to expect near instant response times. Furthermore, as workloads become more parallel, the need for low tail latencies becomes increasingly important since jobs often run at the speed of the slowest request. Second, as workloads become increasingly data-driven, I/O latencies due to storage and networks play a large part in the end-to-end user experience for latency sensitive applications. Storage is often the hardest resource to share and is typically the bottleneck resource. Unless storage can be completely avoided, storage latencies typically have the most impact on overall latency, particularly at the tail. Third, workloads are moving into multi-tenant cloud environments where resources are shared, particularly network and storage. This shift in industry to consolidate workloads onto shared public and private clouds is beneficial in reducing resource and management costs of computing infrastructures. However, while consolidation leads to greater economies of scale, it also introduces challenges in meeting tail latency SLOs. In our proposed work, we will demonstrate that we can build a networked storage system that can meet tail latency SLOs for many workloads sharing the system. Specifically, we will study how to meet tail latency SLOs from the perspective of scheduling policies, admission control, and workload placement/migration. Thesis Committee: Mor Harchol-Balter (Chair)Gregory R. Ganger David G. Andersen Michael A. Kozuch (Intel Labs) Arif Merchant (Google, Inc.) Copy of Thesis Summary

For More Information, Contact:


Thesis Proposal