Product Quick Search

  * Not including RF components
Mercury is contributing its rich SAL API to the developer community for evaluation, use, and augmentation.
» Learn More

Performance Enhanced OpenMPI/OFED

Performance Enhanced OpenMPI/OFED™ from Mercury Computer Systems optimizes Open MPI (the open-source high-performance computing library) with Mercury’s OFED provider software module for serial RapidIO®. OpenMPI/OFED implements the open-standard MPI allocation functions using DMA-optimized buffer allocation methods, resulting in significantly improved application performance. These enhancements combine the productivity and ease-of-use of the MPI API (designed for data-intensive computing) with the speed of an optimized low-latency, high-bandwidth switch fabric (designed for embedded computing). OpenMPI/OFED provides a release of the Open Fabrics Enterprise Distribution (OFED) to serial RapidIO, which enables, for example, high-performance MPI applications to run on Mercury multi-computers. OpenMPI/OFED runs directly on the high-speed fabric, offering an optimized open-source alternative for data-plane communications. By providing an optimized OFED library for serial RapidIO, Mercury has made Open MPI suitable for dense image and signal processing solutions utilizing PPC or Intel® POET™-based processing modules.
  •  MPI integration with ICS for ultimate inner-loop performance 
  • Alternative MPI implementations (open source, commercial) and other OFED-dependent software supported as an optional service 
  • Migration between PowerPC™-based and x86-based processing modules
  • Application portability among multiple systems using serial RapidIO, Ethernet or other high-speed fabrics
  • Optimized memory copy implementations added to Open MPI and OFED for higher performance shared memory communication
  • Higher performance, shared memory MPI communication on serial RapidIO 
  • MPI portability provides code compatible across fabrics 
  • Zero-copy RDMA message-passing support with application buffers 
  • Enhanced MPI allocation provides contiguous memory to enable higher performance communication 
  • RapidIO high throughput and low latency with minimal overhead
System Requirements
Ensemble™ Series platforms with MultiCore Plus software
   Linux OS
      864x PPC Serial RapidIO
      x86 Intel® Serial RapidIO
Privacy Policy