[Colloquium] Rui Liu Candidacy Exam/May 25, 2022

[Megan Woodward]

This is an announcement of Rui Liu's Candidacy Exam.
===============================================
Candidate: Rui Liu

Date: Wednesday, May 25, 2022

Time: 10:30 am CST

Location: JCL 390

Title: Efficient Resource Sharing for Data-Intensive Computation

Abstract: Data-intensive computation has increasingly become prevalent and critical for a wide variety of applications that underpin the modern world. Such computation jobs are usually expensive in terms of time and resource consumption, thus it is beneficial but challenging to efficiently share the resource among them in resource-constrained environments. Our studies show that by exploiting workload characteristics and leveraging resource traits, the constrained resource could be efficiently shared among multiple data-intensive computation jobs to achieve the same performance objective as they only achieve when competing for resources fiercely. In our research, we explore efficient resource sharing for data-intensive computation and implement various prototype systems to realize them. We propose and implement Pack primitive for deep learning training to share common I/O and computing processes among models on the same device. We further propose and design a resource arbitration framework that can continuously prioritize the data-intensive computation jobs and determine if/when to reallocate and preempt the resources for them.

Advisors: Aaron Elmore and Michael Franklin

Committee Members: Aaron Elmore, Michael Franklin, and Sanjay Krishnan


-------------- next part --------------
A non-text attachment was scrubbed...
Name: ruiliu_thesis_proposal.pdf
Type: application/pdf
Size: 2367483 bytes
Desc: ruiliu_thesis_proposal.pdf
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20220517/a17210c0/attachment-0001.pdf>