Initial Operating Capability for the Hypercluster Parallel-Processing Test Bed

The NASA Lewis Research Center is investigating the benefits of parallel processing to applications in computational fluid and structural mechanics. To aid this investigation, NASA Lewis is developing the Hypercluster , a multiarchitecture, parallel-processing test bed. This paper describes the initial opera ting capability (IOC) being developed for the Hypercluster. The IOC will provide a user with a programming/operating environment that is interactive, responsive, and easy to use. The IOC effort includes the development of the Hypercluster operating system (HYCLOPS). HYCLOPS runs in conjunction with a vendorsupplied disk operating system on a front-end processor (FEP) to provide interactive, runtime operations such as program loading, execution, memory editing, and data retrieval. Run-time libraries, that augment the FEP FORTRAN libraries, are being developed to support parallel and vector processing on the Hypercluster. Special utilities are being provided to enable passage of information about application programs and their mapping to the the FEP and the Hypercluster are being handled by dedicated processors, each running a message-passing kernel, (MPK). A sharedmemory interface allows rapid data exchange between HYCLOPS and the communications processors. Input/output handlers are built into the HYCLOPS-MPK interface, eliminating the need for the user to supply separate 110 support programs on the FEP. * operating system. Communications between