Hot Trace Optimizations for Lifeguards

Tuesday Nov. 10, 2009
Hamerschlag Hall D-210
4:00 pm

Olatunji Ruwase
Carnegie Mellon University

Abstract

Lifeguards are run-time correctness checking tools that detect bugs in the execution of unmodified binaries, and are thus valuable for improving the reliability and security of computing systems. The most subtle security, concurrency and memory bugs are detected by lifeguards that monitor at the instruction granularity. Despite their advantages, lifeguards are rarely deployed in production settings because instruction-grained monitoring is too slow.

In this talk, I will present results from our studies on how to accelerate instruction-grained lifeguards. We observed that raising the granularity of monitoring to dynamic program paths (traces) exposes significant redun-dancies in a variety of instruction-grained lifeguards without weakening their bug detection effectiveness. Evaluations of our proposed software optimizations on state of the art soft-ware(Valgrind) and hardware(LBA) lifeguard support platforms showed lifeguard speedups of up to 30% on Valgrind and 2X on LBA.

Bio

Olatunji Ruwase is a Computer Science Ph.D. student at Carnegie Mellon University. Under the tutelage of Prof. Todd Mowry and in the context of the Log Based Architectures(LBA) project, Olatunji is studying novel software and hardware techniques for accelerating dynamic correctness checking tools.

Back to the seminar page