GPI - Efficient and scalable multicore programming
The advantages of GPI are the communication performance at wire speed and its scalability on multicore systems. Several benchmarks demonstrate the excellent scalability of GPI, also in cases where MPI scales poorly.
GPI directly extends the simple threadbased programming model from the node level to the cluster level. GPI works with Posix threads or with the optimized hardwareaware multicore thread package MCTP.
GPI handles all communication asynchronously without interrupting the CPU and allows perfect overlapping with computation.
GPI is lightweight and easy to use and includes additional functionality like fast barriers, collective operations or atomic counters. Together with the simple execution model and the extensive environment checks, GPI with MCTP is the choice to program very large systems.
GPI was developed by Fraunhofer ITWM and has already successfully been used in various industrial application codes. Now GPI and MCTP are available as licensed products for C, C++ and Fortran under Linux through ITWM's marketing partner scapos. At the ISC'10 exhibition it will be demonstrated at the ITWM booth.
The Fraunhofer Competence Center for HPC at the Fraunhofer Institute for Industrial Mathematics (ITWM) in Kaiserslautern (http://www.itwm.de/de/hpc/hpc/) offers innovative HPC solutions for the industry and the HPC market. Their R&D is focused on HPC tools, service oriented computing, and server based visualization.
Scapos (www.scapos.com) is a spinoff company and marketing partner of Fraunhofer. Scapos distributes various software products in the technical computing and HPC market. Contact: email@example.com