PhD defence - Pansy Arafa

Tuesday, May 23, 2017 1:00 pm - 1:00 pm EDT (GMT -04:00)

Candidate

Pansy Arafa

Title

Heterogeneous Resource Leveraging in Vehicular Networks

Supervisor

Sebastian Fischmeister

Abstract

The complexity of modern software systems has been rapidly increasing. Program debugging and testing are essential to ensure the correctness of such systems. Program analysis is critical for understanding system’s behavior and analyzing performance. Many program analysis tools use instrumentation to extract required information at run time. Instrumentation naturally alters a program’s timing properties and causes perturbation to the program under analysis. Soft real-time systems must fulfill timing constraints. Missing deadlines in a soft real-time system causes performance degradation. Thus, time-sensitive systems require specialized program analysis tools. Time-aware instrumentation preserves the logical correctness of a program and respects its timing constraints. Current approaches for time-aware instrumentation rely on static source-code instrumentation techniques. While these approaches are sound and effective, the need for running worst-case execution time (WCET) analysis pre- and post-instrumentation reduces the applicability to only hard real-time systems where WCET analysis is common. They become impractical beyond microcontroller code for instrumenting large programs along with all their library dependencies. In this thesis, we introduce theory, method, and tools for time-aware dynamic instrumentation realized in DIME tool. DIME is a time-aware dynamic binary instrumentation framework that adds an adjustable bound on the timing overhead to the program under analysis. DIME also attempts to increase instrumentation coverage by ignoring redundant tracing information. We study parameter tuning of DIME to minimize runtime overhead and maximize instrumentation coverage. Finally, we propose a method and a tool to instrument software systems with quality of service (QoS) requirements. In this case, DIME collects QoS feedback from the system under analysis to respect user-defined performance constraints. As a tool for instrumenting soft real-time applications, DIME is practical, scalable, and supports multi-threaded applications. We present several case studies of DIME instrumenting large and complex applications such as web servers, media players, control applications, and database management systems. DIME limits the instrumentation overhead of dynamic instrumentation while achieving a high instrumentation coverage.