Responsive Parallel Computation

Nearly all parallel computing research focuses on compute-intensive applications drawn from areas such as scientific computing, optimization, and machine learning and AI.  The goal is to execute a single parallel job to completion as quickly as possible my maximizing throughput.  With the advances in multicore chips, the application domain of parallelism has grown to include interactive applications that demand responsiveness.  Such applications, such as games, servers, robotic systems, consists of many tasks some of which interact with the external world (users, clients, obstacles, etc.) and require quick response, while some other tasks run in the background.  The goal of this project is to develop the underlying principles, including the programming theory and algorithms, on which such responsive parallel systems could be developed. 


ACM SIGPLAN Conference on Programming Language Design and Implementation (PLDI). 2017.
Stefan K. Muller, Umut A. Acar, Robert Harper.

Latency-Hiding Work Stealing.ACM Symposium on Parallel Algorithms and Architectures (SPAA). 2016.
Stefan K. Muller and Umut A. Acar.
The full version can be found here.