[Colloquium] Ruidan Li MS Presentation/Aug 17, 2023

[Megan Woodward]

This is an announcement of Ruidan Li's MS Presentation
===============================================
Candidate: Ruidan Li

Date: Thursday, August 17, 2023

Time: 10 am CST

Location: JCL 298

Remote Location: https://uchicago.zoom.us/j/98464427616?pwd=b0I2ZENuemN0UEFGcnNpVUNzMXVXZz09&from=addon

M.S. Paper Title: FREP: Fleet Performance Evaluation Pipeline for Identifying Server Degradation in Large-Scale Datacenters

Abstract: Stable and predictable performance of fleet servers is critical for providing reliable services. To identify and remove underperforming servers, performance data must be recorded and then systematically analyzed at scale. In this thesis, we report on our methodology for collecting and analyzing performance data at CompanyX for the purposes of identifying and removing underperforming datacenter servers. We present (and will release), to the best of our knowledge, the largest dataset on CPU and memory benchmarks. We show how off-the-shelf statistical techniques must be customized for the nature of our dataset, resulting in a custom pipeline, FREP, that leverages a wide range of statistical techniques. FREP comprises a variety of modules that detect variances, divergences, and outliers, and has provided various types of important recommendations to CompanyX operators and helped improve their understanding of their fleet behaviors.

Advisors: Haryadi Gunawi

Committee Members: Haryadi Gunawi, Hank Hoffmann, and Marty Humphrey


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20230803/3b979256/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ruidan_li_master_thesis.pdf
Type: application/pdf
Size: 5209578 bytes
Desc: ruidan_li_master_thesis.pdf
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20230803/3b979256/attachment-0001.pdf>