Junfeng Yang Awarded NSF Grant to Address Concurrency Errors
LOOM: a Language and System for Bypassing and Diagnosing Concurrency Errors.
This project addresses programming challenges posed by the new trend in multicore computing. Multithreaded programs are difficult to write, test, and debug. They often contain numerous insidious concurrency errors, including data races, atomicity violations, and order violations, which we broadly define to be races. A good deal of prior research has focused on race detection. However, little progress has been made to help developers fix races because existing systems for fixing races work only with a small, fixed set of race patterns and, for the most part, do not work with simple order violations, a common type of concurrency errors.
The research objective of this project, LOOM: a Language and System for Bypassing and Diagnosing Concurrency Errors, is to create effective systems and technologies to help developers fix races. A preliminary study revealed a key challenge yet to be addressed on fixing races that is, how to help developers immediately protect deployed programs from known races. Even with the correct diagnosis of a race, fixing this race in a deployed program is complicated and time consuming. This delay leaves large vulnerability windows potentially compromising reliability and security.
To address these challenges, the LOOM project is creating an intuitive, expressive synchronization language and a system called LOOM for bypassing races in live programs. The language enables developers to write declarative, succinct execution filters to describe their synchronization intents on code. To fix races, LOOM installs these filters in live programs for immediate protection against races, until a software update is available and the program can be restarted.
The greatest impact of this project will be a new, effective language and system and novel technologies to improve the reliability of multithreaded program, benefiting business, government, and individuals.