Applying Software Testing for Hardware Resiliency Analysis

Wednesday, May 03, 2017
12:00 PM to 1:00 PM
UTC 2.112A
Free and open to the public

This talk will present a project that bridges some research on software and hardware; time permitting, we will also discuss pros and cons of collaborating across areas. As hardware technology scales, efficient resiliency solutions are needed to address the increased likelihood of hardware errors. Silent data corruptions (SDCs) are especially harmful because they are not easily recognizable. Several recently proposed analyses have used error injections to identify SDC-causing instructions, but these analyses remain too expensive for practical use or sacrifice accuracy to improve performance.  We show how applying software testing techniques -- including test coverage, test minimization, and test prioritization -- can advance the state-of-the-art resiliency analysis, improving both accuracy and performance of SDC discovery. Our experiments show that using our approach can speed up SDC discovery up to 10x, while uncovering more SDC-causing instructions.

x x



Darko Marinov

University of Illinois at Urbana-Champaign

Darko Marinov is an Associate Professor in the Department of Computer Science at the University of Illinois at Urbana-Champaign.

His main research interests are in Software Engineering, especially improving software quality using software testing.  He has a lot of fun finding and preventing software bugs.  His work has been supported by Google, IBM, Intel, Microsoft, NSF, Qualcomm, and Samsung.