Advanced Parallel Programming Models

[ Overview | Description ]

    The advanced parallel programming model research at CAPSL is an investigation of potentially revolutionary methods of HPC programming, not derived incrementally from current conventional practices. Our effort seeks to explore and develop new programming models whose purpose is to provide more effective ways for programmers to express algorithms on high performance computing platforms, addressing many of the difficult issues associated with issues such as scalability, memory hierarchies, and memory and functional unit latencies, for both regular and irregular problems. Our focus has been parallel programming models for large scale parallel machines, including future generation high-end parallel architectures.

    Advanced programming model concepts currently under investigation at CAPSL include (but are not restricted to)

  1. percolation model;
  2. location consistency and atomicity;
  3. asynchronous threads and their synchronization based on (1) and (2).

    We are interested not only in the semantics of programming concepts but also, and most importantly, their efficient implementation. To this end, new compilation models and runtime systems are being proposed, investigated and implemented both on conventional platforms and multithreaded architectures. Compilation model and runtime systems for parallel machines are closely related to the program execution models of the target architectures. To this end, Threaded-C has been a good case study as a specification of the EARTH Virtual Machine (EVM) and has been an important link to out study high-level programming models and system software for multithreaded architectures. We are currently studying new execution models for the efficient mapping of the advanced programming model concepts as mentioned earlier.



© CAPSL 1996-2013. All Rights Reserved.