As we inch closer to exascale HPC systems, it is necessary to consider new models of execution which efficiently utilize the massive resource pool of these systems and exploit the potential billion-way parallelism of exascale applications. ParalleX is an explorative execution model proposed to achieve exaFlop performance by addressing inefficiencies inherent to convention models of execution. In particular, ParalleX implements an asynchronous, message-driven execution environment to mask the latency of accessing parallel resources and utilizes a global address space to simplify parallel programming and facilitate dynamic load balancing. In an attempt to address the longstanding I/O bottleneck of computer systems, ParalleX requires the design of a new I/O subsystem which closely matches the semantics of its execution model. We detail the design of an I/O library suitable for integration with the ParalleX model and address future research necessary in developing PXFS.

Bio:
Shane Snyder recently received his M.S. in Computer Engineering from Clemson University, where most of his research revolved around the OrangeFS parallel file system and parallel I/O. Some of Shane's other research interests include exascale HPC, parallel programming and execution models, and parallel applications.