ParaStation MPI provides a standard interface for parallel applications that require either MPI or MPI-2 library functionality. ParaStation MPI has been design to select the most appropriate of all available interconnects at runtime. For example, for intra-node communication, ParaStation MPI will select a the shared memory model (shmem). Inter-node communication is done using the native networking libraries of the highest performing installed interconnect (Infiniband, Myrinet etc..).
In June 2009, ParaStation MPI helped to propel the 3288 node JuRoPA cluster at the Jülich Supercomputer Centre to an impressive 274,8 Teraflops of sustained performance over an 11 hr period and attained an overall parallel efficiency of 91,6%. At that time JuRoPA Ranked No. 10 in the World and No. 1 in Europe of the most powerful general purpose supercomputers (top500.org). This benchmark run demonstrated the unsurpassed scalability of the ParaStation MPI - which was shown to scale to more than 25,000 MPI processes (without threading) on this machine.