Nick Mitchell and Glenn Ammon talk Dec. 17

[Sheila D Clark]

Seminar
Friday, December 17
4pm, 4405 SC

Dr. Nick Mitchell and Dr. Glenn Ammon
IBM T.J. Watson Research Center

"Bottlenecks and Leakbot: Two tools for debugging enterprise size
applications" 

Enterprise applications are unlike most programs you know.  These 
applications are built by hundreds of programmers, who adapt and patch 
together modules from dozens of vendors.  The result is an application 
that no one understands. 

Diagnosing problems in these applications is hard, particularly for 
global problems like poor performance and high memory usage.  These 
problems often arise after development, sometimes in systems that are 
critical to a business.  A forensic approach to program diagnosis is 
to run the application and analyze its leavings.  In this talk, we 
describe Leakbot and Bottlenecks, which are two lightweight instances 
of this approach. 

The leavings are large, because these are large programs.  However, 
the 80-20 rule applies, and the problem can be described succinctly. 
The trick is finding that succinct description.  Bottlenecks is an 
interactive tool that helps an analyst iteratively refine a good 
description of a performance problem.  By contrast, Leakbot 
automatically infers structure in a heap, assigns heap objects to 
memory regions, and concisely reports regions that appear to be memory 
leaks. 


Bios:
Glenn Ammons received his doctorate from the University of Wisconsin in
2003, under Rastislav Bodik. He is currently a research staff member at
IBM's Watson Research Center, where he works on improving software
development by analyzing program executions. 

Nick Mitchell received his Ph.D. from the University of California - San
Diego in 2000, under Jeanne Ferrante and Larry Carter. His graduate research
focused on performance prediction and architectural models for scientific
applications. Since late 2000, Nick has been a research staff member at
IBM's Watson Research Center, where his research has focused on performance
and memory analysis of very large applications in a "production" setting.