Guest Speaker

Achieving Robust Software Systems After Deployment

Michael Bond
Dept. of Computer Sciences
University of Texas at Austin

Feb 2 2010 3:30PM
480 Dreese Labs
All interested parties are invited to attend.
Refreshments will be served prior to talk.

Abstract:

Society relies increasingly on software to improve science, health care, communications, education, and energy. Unfortunately software is becoming more complex and concurrent due to feature demand and hardware trends that are leading to more instead of faster cores. In the face of these challenges, developers have trouble writing large, correct, scalable programs. All deployed software contains bugs, even if it has been thoroughly tested, because it is infeasible to test all possible inputs, environments, and thread schedules in large systems. My research focuses on improving reliability while production software runs, to help prevent, diagnose, and tolerate errors that actually manifest in deployment.

This talk first presents Pacer, a deployable, scalable approach for detecting data races, which are a common and serious type of concurrency bug. Pacer detects every race that occurs with a probability equal to the sampling rate with time and space overheads proportional to the sampling rate. Second, I describe Breadcrumbs, which efficiently reports the calling context (stack trace) of concurrency and other bugs -- essential information for understanding the behavior of complex, modern programs. Breadcrumbs adds instrumentation that computes probabilistically unique values continuously for each calling context, and later performs a backward heuristic search to reconstruct the contexts of reported bugs. I conclude with my future plans for making concurrent software more robust, scalable, and secure.

Bio:
Michael D. Bond is a postdoctoral fellow in Computer Science at UT Austin. He received his PhD from UT Austin in December 2008, supervised by Kathryn S. McKinley. His research makes software more robust by using deployable dynamic analysis to prevent, diagnose, and tolerate unexpected errors. His interests include programming languages, runtime systems, compilers, and security. His dissertation received the ACM SIGPLAN Outstanding Doctoral Dissertation Award in 2009.

Host: Atanas Rountev

* Michael Bond is a CSE faculty candidate