Overview
This work seeks to better understand the behavior of soft real-time applications and provide tools for the programmer to maximize the quality of service provided by such applications (maximizing the computational quality while respecting the soft real-time deadlines.
The first part of this work explored how to analyze an application using a profile-driven approach to identify the most variant parts of an application which are most likely to cause problems for soft real-time constraints. In a second part, we looked at how for certain applications that have variable semantics, an automated system could dynamically tune an application to maximize the quality of the computation while maintaining soft real-time deadlines.
This work is led by Tushar Kumar.
Papers
The dynamic tuning aspect of this work was presented at EMSOFT 2010. The paper is available and the slides are available as a PowerPoint file or a PDF file.
The FSE work was followed by a paper in LCPC 2008.
Work on identifying the most variant parts of an application was presented in FSE 2007 as a short paper with an associated poster.
