CS Events

Abstract: 

Scaling stateful applications with available core counts is not a straightforward task. As core counts increase, shared data structures and hardware resources become bottlenecks and limit potential performance improvements. My work aims to utilize thread and data scheduling to overcome these potential bottlenecks. In this talk, I will first discuss my previous work Pfsck, where we parallelize a local file-system checker and utilize dynamic thread scheduling to maximize performance for varying filesystem characteristics. I will follow up with one of my ongoing works, Kamino, where we utilize request scheduling to scale a distributed VM allocator and improve latencies and memory efficiency for varying traffic loads and request distributions.