[Colloquium] Chen Zou Candidacy Exam/Dec 9, 2021

meganwoodward at uchicago.edu meganwoodward at uchicago.edu
Wed Dec 8 08:37:46 CST 2021


This is an announcement of Chen Zou's Candidacy Exam.
===============================================
Candidate: Chen Zou

Date: Thursday, December 09, 2021

Time: 10 am CST

Remote Location: https://uchicago.zoom.us/j/99950863095?pwd=Tk1TNHJEK04vSk9yeTNWZU5yb2xvZz09 Meeting ID: 999 5086 3095 Passcode: 310501

Location: JCL 298

Title: Architecting computational storage systems in data centers

Abstract: The past decades have seen explosive growth in large-scale data which boosts the transition in data centers from hard disk drives to solid-state drives for storage performance while storing and processing these data. Driven by the continuously improving flash bandwidth and the compute demands to match this bandwidth, the question of how to best divide computation across compute and storage resources is actively being revisited. Computational storage, the idea of enabling some computation inside storage devices to early filter or late increment data for reduced interconnect traffic and transfer time resurfaces in both storage and computing community.

However, existing research mostly conducts point studies on reporting how a single application would benefit from offloading certain functions to the storage device(s). There is little general knowledge on how a computational storage system should be architected to efficiently provide general and high-performance acceleration that can match the growing flash bandwidth.

In this thesis project, we try to answer three questions around how to architect a computational storage system that provides general support for different workloads. First, what kinds of acceleration opportunities are there for different applications? Second, what kinds of SSD architecture provide efficient support for computational storage? Third, what are different computational storage processor architecture elements that enable high performance, matching or exceeding continuously improving flash bandwidths?

Advisors: Andrew Chien

Committee Members: Andrew Chien, Yanjing Li, and Hank Hoffmann



More information about the Colloquium mailing list