Panel

Moderator

Hartmut Kaiser, Center for Computation & Technology at Louisiana State University, USA
Laxmikant (Sanjay) Kale, Department of Computer Science, University of Illinois at Urbana-Champaign, USA
Martin Berzins, Scientific Computing and Imaging Institute, University of Utah, USA
Mike Bauer, NVIDIA, USA
Thomas Fahringer, University of Innsbruck, Austria

Why should one choose to use AMT models?
What kinds of applications will benefit from using AMT models and programming systems?
For what kinds of applications is AMT-style programming better suited to creating maintainable code than message passing-style code?
Are AMT programs more difficult to debug because it is (1) easier to create race conditions and (2) more difficult to get anything like a stack trace? What is being done to make debugging of large-scale AMT applications easier?
What sort of language support is needed for AMT programming?
- What advantage do newer languages like Go, Julia, etc., have for AMT programming, and can/should they displace C++ for scientific applications?
- Many languages have added “await” or some variant as new keywords for processing asynchronous code. How important is this feature for the success of the AMT paradigm?
How does one choose which AMT programming model among others?