CAREER: A Task-Parallel Platform for Interactive Applications

Washington University, Saint Louis MO

Investigators

Abstract

Modern software tends to be service-oriented, long running, and interactive. The interaction can be external, such as waiting on a user input or a network package, or internal, where the components that make up the application interact with one another, often asynchronously. It is important that interactive applications run fast on commodity multicore hardware, because fast interactions lead to good user experiences, and multicore processors are the bargain components of today's hardware technology. Fast interactive parallel applications can be challenging to develop, however, as the programmer must deal with multiple non-trivial concerns simultaneously, including asynchrony, concurrency, and ensuring application responsiveness. This project focuses on developing software infrastructure to enable non-expert programmers to write parallel interactive applications that run fast on multicore hardware. The resulting platform provides: (1) a high-level language abstraction that decouple parallelism and concurrency specifications from scheduling and encapsulate nondeterminism; (2) a provably and practically efficient scheduler that achieve various desirable performance criteria; (3) a well-defined cost model to reason about the performance of the computation; and (4) a set of efficient tools to debug code written using the abstraction. The resulting platform enables programmer productivity in developing parallel interactive applications and efficient utilization of multicore hardware. The research results from this project is disseminated broadly though top parallel computing conferences. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.