Brief Overview

The goal of the ParalleX (PX) model of computation is to address the key challenges of efficiency, scalability, sustained performance, and powe consumption with respect to the limitations of conventional programming practices (e.g., MPI). ParalleX will improve efficiency by reducing the average synchronization and scheduling overhead, improve utilization through asynchrony of work flow, and employ adaptive scheduling and routing to mitigate contention (e.g., memory bank conflicts). Scalability will be dramatically increased, at least for certain classes of problems, through data directed computing using message-driven computation and lightweight synchronization mechanisms that will exploit the parallelism intrinsic to dynamic directed graphs through their meta-data. As a consequence sustained performance will be dramatically improved both in absolute terms through extended scalability for those applications currently constrained and in relative terms due to enhanced efficiency achieved. Finally, power reductions will be achieved by reducing extraneous calculations and data movements. Speculative execution and  speculative prefetching are largely eliminated while dynamic adaptive methods and multithreading in combination serve many of the purposes these conventionally provide.

For more information, please visit our website