Undergrad Research Project - Simple architectural support for latency-critical applications

Spring 2018

Amolak Nagi
Nathan Beckmann
Project description

Datacenter applications distribute work across hundreds of nodes, with the end-to-end completion time determined by the slowest node. This makes them sensitive to tail latency, not throughput, for performance. Recent work has proposed a number of sophisticated mechanisms to improve tail response time, but these depend upon complex architectural support that is not present in current servers. This project will investigate much simpler, and therefore practical, mechanisms to improve tail latency. Specifically, we will begin by re-designing Ubik, a cache partitioning technique, to use the caching mechanisms already present in current-generation Intel processors.

Return to project list