Seminars

DATE: Thursday, November 8, 2018
TIME: 12:00 - 1:00 pm
PLACE: RMCIC Panther Hollow Conference Room, 4th Floor

SPEAKER: Nosayba El-Sayed, Dept of Computer Science, Emory College

TITLE: Why do we need data-driven datacenters?

ABSTRACT:
Designing datacenters that are reliable, energy-efficient, and capable of delivering high performance and high utilization is a nontrivial mission. In this talk, I argue that we need more data-driven operations in datacenters to better optimize these design goals. I will present three examples that demonstrate how data-driven predictive modeling can help us anticipate and mitigate critical events in datacenters such as job failures [ICDCS’17], database server load changes [SIGMOD’18], and even changes in the demands of co-located jobs [HPCA’18]. I will discuss how these different predictions can improve the way we allocate and manage resources dynamically in datacenters. Finally, I will present a recent, open-sourced tool that leverages clustering techniques to unlock significant performance on modern datacenter hardware through lightweight cache profiling and partitioning.

BIO:
Nosayba El-Sayed is a new lecture-track faculty at the Computer Science department in Emory University. Before joining Emory, she was a Postdoctoral Associate at the Computer Science and Artificial Intelligence Lab (CSAIL) in MIT. Nosayba's research focuses on designing data-driven techniques that exploit the wealth of data generated in modern datacenters. Nosayba completed her PhD at the University of Toronto under the supervision of Bianca Schroeder, during which time she interned at Amazon's Datacenter Global Services division to design server-outage prediction mechanisms. Nosayba's work has been published in venues such as SIGMETRICS, DSN, ICDCS, SC, SIGMOD and HPCA. Her work on datacenter reliability was featured in ;login! Usenix Magazine, Data Center Knowledge, and Communications of the ACM.

SEMINAR HOST: George Amvrosiadis

SDI / ISTC SEMINAR QUESTIONS?
Karen Lindenfelser, 86716, or visit www.pdl.cmu.edu/SDI/