``Automatic software fault diagnosis by exploiting application signatures"

Xiaoning Ding, Hai Huang, Yaoping Ruan, Anees Shaikh, and Xiaodong Zhang

Proceedings of 22nd USENIX Conference on Large Installation System
Administration, (LISA'08), San Diego, California, November 9-14, 2008.


Abstract

Application problem diagnosis in complex enterprise environments is a
challenging problem, and contributes significantly to the growth in IT
management costs.  While application problems have a large number of
possible causes, failures due to runtime interactions with the system
environment (e.g., configuration files, resource limitations, access
permissions) are one of the most common categories. Troubleshooting
these problems requires extensive experience and time, and is very
difficult to automate.  In this paper, we propose a black-box approach
that can automatically diagnose several classes of application faults
using applications' runtime behaviors.  These behaviors along with
various system states are combined to create signatures that serve as a
baseline of normal behavior.  When an application fails, the faulty
behavior is analyzed against the signature to identify deviations from
expected behavior and likely cause.  We implement a diagnostic tool
based on this approach and demonstrate its effectiveness in a number of
case studies with realistic problems in widely-used applications. We
also conduct a number of experiments to show that the impact of the
diagnostic tool on application performance (with some modifications of
platform tracing facilities), as well as storage requirements for
signatures, are both reasonably low.

Back to the HPCS Main Page at the Ohio State University.