Columbia University
Events Banner
VMware Tech Talk

Date:February 27, 2013 from 6:00 pm to 7:30 pm EST
Location:CESPR, Davis Auditorium
Contact:For further information regarding this event, please contact Center For Career Education by sending email to .

VMware Tech Talk
Wednesday, February 27th @ 6pm

Title: From Automatic Performance Diagnosis to Automatic Parallel/Distributed Execution of Machine Learning (ML) Tasks 

There are many opportunities to apply techniques from statistical machine learning to problems in systems and vice versa (to apply systems ideas to aspects of statistical machine learning). In this talk I provide examples from case-studies inside VMware.

First, I present vPerfGuard, a system for automatically explaining and diagnosing application performance issues via online modeling and the datamining of datacenter telemetry. In addition to model construction, vPerfguard employs statistical techniques for online model-management, specifically to cope with changing conditions in the datacenter (and/or application workloads) that cause the constructed models to drift (lose their explanatory/predictive power over time).

Second, I describe MLBase, an early-stage collaboration between VMware and the AMP Lab at U.C. Berkeley. MLBase is a system for efficiently executing machine learning tasks in a distributed environment/cluster. At the heart of MLBase is a novel optimizer that borrows ideas from database query planning and optimization to identify efficient execution plans for machine learning tasks while codifying some of best practices of machine learning experts.

Both vPerfguard and MLBase will appear in ICPE 2013 and CIDR 2013 respectively.


Rean Griffith is a Staff Engineer in the CTO's office at VMware. Prior to joining VMware in 2010, he was a post-doc in the RAD Lab at U.C. Berkeley. He received his M.Sc. and Ph.D. in Computer Science from Columbia University in 2003 and 2008 respectively, and a B.Sc. in Computer Science and Management from the University of the West Indies (Cavehill, Barbados) in 2000. His research interests include distributed systems, operating systems, adaptive systems and networks, control systems, performance and reliability modeling, and the application of statistical machine learning to resource management and systems problems.

Add To My Calendar