Supercomputing and Computation

SHARE

Asynchronous Speculative Methods for Scientific and Mathematical Computing


Problem Statement:

  • Scalability is severely impaired with traditional approaches in supercomputing applications. To enable exascale computation, new technology is needed to relieve synchronization overheads.  Moreover, the solution must be evolutionary to transparently port existing applications to exascale platforms.

Technical Approach:

  • Global synchronization latencies are hidden to remove idle time via speculative execution. Rollback methods are used to recover from synchronization errors. A transparent speculative interface is added as backward-compatible extensions to Message Passing Interface (MPI) standard.

Benefit:

  • The utilization of large-scale parallel executions increases on very large scale systems.  The performance/price ratio is improved &  turnaround time for simulations is greatly reduced for scientists using the large scale executions.
ASK ORNL

We're always happy to get feedback from our users. Please use the Comments form to send us your comments, questions, and observations.