Wednesday, February 12

MS18
Exploiting Task Parallelism in Exascale Computing Era

3:10 PM - 4:50 PM
Room: 605

As we approach the age of exascale computing, developers of large scale scientific applications will be facing new challenges in terms of performance and portability. In this regard, task parallel computation paradigm presents a promising way forward as it allows programmers to express a high degree of parallelism, utilize advanced scheduling techniques based on the data dependency information available, and obtain architectural portability relatively easily through runtime support. This minisymposium will feature talks on recent developments in task-parallel runtimes, middlewares and applications to explore the advantages of the task-parallel computation paradigm, as well as to identify the challenges ahead for a successful transition to the exascale era.

Organizer: H. Metin Aktulga
Michigan State University, U.S.
Umit V. Catalyurek
Georgia Institute of Technology, U.S.

3:10-3:30 DeepSparse: A Task-Parallel Framework for Sparse Solvers on Deep Memory Architectures abstract

Md Afibuzzaman and Fazlay Rabbi, Michigan State University, U.S.; Yusuf Ozkaya and Umit V. Catalyurek, Georgia Institute of Technology, U.S.; H. Metin Aktulga, Michigan State University, U.S.

3:35-3:55 Exploiting Task Parallelism in an Exascale Ecosystem: Some Issues and Possible Solutions abstract

Olivier Aumage, Inria, France

4:00-4:20 Novel Approaches to Optimize and Execute Task-Based, Irregular Applications on Extreme-Scale, Heterogeneous Systems using PaRSEC abstract

Thomas Herault, University of Tennessee, Knoxville, U.S.

4:25-4:45 Asynchronous Programming in Modern C++: What Is An AMT and Why Do You Want One for Christmas? abstract

Hartmut Kaiser, Louisiana State University, U.S.