Wednesday, June 18, 2014
10:00 AM
CSB Room 703
Hans Vandierendonck
Queen's University Belfast, UK
Parallel Programming using a Task Dataflow Specification
Task dataflow-style execution models assume that the programmer (in-)directly specifies parallel programs as directed acyclic graphs of tasks (DAGs). These DAGs are executed in parallel and scheduled on multiple processing cores automatically, much like a superscalar processor would dynamically schedule instructions on execution units. This frees the programmer from explicitly identifying parallelism, and from solving many low-level problems related to scheduling and synchronization.

In this talk, we will present the motivation for our interest in this class of programming models and we will discuss some opportunities where task dataflow models can bring new advances. We will discuss our initial results on the Swan runtime system supporting both recursive and task dataflow parallelism. Swan extends the Cilk language and scheduler, uses managed objects to ease inter-task dependence tracking and uses a low-overhead mechanism for inter-task dependence tracking reminiscent of ticket locks. Through its design, it shows good scalability and reaches linear speedup when tasks execute for 200 usec or longer on a 48-core machine.

BIO: Hans Vandierendonck is assistant professor at Queen's University Belfast, UK, in high-performance and distributed computing. He graduated from and obtained the PhD degree from Ghent University, Belgium. Hans was awarded several prizes, among which the IBM Belgium Prize for Informatics twice, a rare privilege, once in 2000 for his graduation thesis and once in 2004 for his PhD dissertation. His main research interests are in parallel computing, runtime systems and compilation. He also has a vested interest in computer architecture. Hans holds an EC Marie Curie Fellowship. He is a Senior Member of the IEEE and a Senior Member of the ACM.