Title: Managing Contention in Large-Scale Systems

Speaker: Lloyd Brown

Advisors: Scott Shenker, Sylvia Ratnasamy

Date: Friday, December 12, 2025

Time: 10AM-11AM PT

Abstract: Contention breeds interesting problems in large-scale systems. In this talk we will discuss two problem domains, congestion control and cluster management systems. In relation to congestion control we focus on the wide-area and ask, from first principles, what the ideal properties are for bandwidth allocations under congestion and find that TCP-friendliness, the working assumption for the last 20+ years, is inconsistent with these properties. In relation to cluster management systems we explore whether general purpose checkpointing can be used to improve the completion rate of low priority jobs, finding that the state of the art periodic checkpointing scheme from HPC is suboptimal.