 |
RadarDataflow Analysis for Concurrent Programs using Datarace DetectionFunded by NSF Award CCF-0811512 |
 |
Overview
Advances in static algorithms for program optimization and error detection have shown that compiler technology can dramatically improve the reliability and performance of computer systems. However, most of these algorithmic advances are limited to sequential programs and ignore the challenges introduced by concurrency, where the need for static checking and potential for optimization are the greatest.
The broad goal of Radar is to develop scalable and precise analysis techniques for concurrent programs. Achieving scalability and precision at the same time, however, is difficult because there is a widely acknowledged tension between these two goals. To address this challenge, our general methodology will be to design analyses that are sound and scalable first (likely at the expense of precision), and then to iteratively refine these analyses by empirically identifying the common concurrent programming idioms where our analyses lose precision, and developing analysis techniques targeted at these idioms.
Our research will lead to the following contributions:
- Radar framework. First, we will design, implement and evaluate a framework called Radar for automatically generating precise concurrent analyses from their sequential counterparts. This will make concurrent analyses much easier to write, allowing compiler writers and analysis writers to easily adapt current analyses to account for concurrency.
- Improved software reliability. Second, because our research plan involves testing and evaluating Radar on a variety of analyses that are used in static checkers, our research will result in sound and scalable checkers that are precise in the face of concurrency. This in turn will lead to improved software reliability.
- Empirical Understanding of Concurrency Patterns. Third, by examining where our initially imprecise (but sound and scalable) Radar framework loses precision, we will identify and categorize common concurrency patterns observed in real systems. This concurrency catalog of common idioms will be useful beyond the confines of our research project, for example in guiding the design of new languages or programming models for concurrency, and in helping researchers and practitioners focus their efforts on building software engineering tools for concurrency.
Publications
- Dataflow Analysis for Concurrent
Programs using Datarace Detection (PLDI 2008)
Ravi Chugh, Jan Voung, Ranjit Jhala, and Sorin Lerner -
RELAY: Static Race Detection on Millions of Lines
of Code (ESEC/FSE 2007)
with Jan Voung, Ranjit Jhala, and Sorin Lerner