Reconfirming the Workshop

Final Report

In the high performance computing world, new architectures (e.g., PIM, multithread) and new applications (e.g., graph theoretic, multi-agent, data mining) are emerging that challenge the status quo of application development and programming languages. Billions of dollars have been invested in applications written in C or Fortran that accomplish parallel efficiency through MPI. The challenge is to leverage new architectures, enable new applications, safeguard existing investments, and seek greater productivity through advances in programming languages.

This workshop will examine key applications in the context of emerging architectures and advances in modern programming languages and runtime systems. The outcome of the workshop will be a plan for HPC that can express algorithms not well suited to the MPI/MPP model of parallel computation. While a revolutionary development in programming languages and applications would be ideal, an evolutionary path offering improved efficiency will also be needed.

Five core areas will be addressed:

1. Supporting applications that exhibit radically different characteristics (particularly in terms of memory access patterns);
2. Enhancing programmer "productivity";
3. Expressing algorithms in environments where the formulation of the solution changes as problem size scales;
4. Scaling systems to trans-peta sizes (and beyond!); and
5. Exploiting the features provided by novel architectures.

To discover emerging requirements, applications will be examined from the perspective of the developer's difficulty expressing them in MPI (which impacts productivity because the algorithms are difficult to write "intuitively", or high performance is extremely difficult to obtain), the computer architect's ability to support the applications requirements (in traditional and emerging machines), and the programming language and runtime system writer's ability to bridge the gap. The output of this effort is feedback to the HPC language community, expressing the needs of HPC users that are not currently met, as well as defining a set of important applications that may benefit from novel architectures.

Your ideas and participation in making this workshop a success are needed.