Enhanced Support for OpenSHMEM Communication in Portals
Authors:
- Brian W. BarrettBBBrian W. Barrett
- Ron BrigthwellRBRon Brigthwell
- K. Scott HemmertKHK. Scott Hemmert
- Kevin PedrettiKPKevin Pedretti
- Kyle WheelerKWKyle Wheeler
- Keith D. UnderwoodKUKeith D. Underwood
- 1 August 2011
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 61-69
Abstract
The looming challenges of exascale computing are generating renewed interest in the Partitioned Global Address Space (PGAS) parallel programming model. From a high-performance interconnect perspective, the one-sided communication model in PGAS addresses some of the challenges that could inhibit the viability of traditional two-sided MPI message passing on millions of cores. The OpenSHMEM project is a recent effort at standardizing a lightweight one-sided communication interface to enable the development of highly scalable PGAS applications. The Portals data movement layer is a low-level one-sided interface that has successfully supported several higher-level one-sided and two-sided interfaces, including PGAS network transport layers. This paper describes the design and implementation of OpenSHMEM on the latest generation of the Portals interface, including a description of how Portals has evolved to maximize performance for PGAS-style communication operations.Keywords
This publication has 12 references indexed in Scilit:
- Enabling Flexible Collective Communication Offload with Triggered Operations2011
- Introducing OpenSHMEM2010
- Evaluation of Remote Memory Access Communication on the Cray XT32007
- SeaStar Interconnect: Balanced Bandwidth for Scalable PerformanceIEEE Micro, 2006
- Design, Implementation, and Performance of MPI on Portals 3.0The International Journal of High Performance Computing Applications, 2003
- The Quadrics network: high-performance clustering technologyIEEE Micro, 2002
- SCI: Scalable Coherent Interface1999
- Co-array Fortran for parallel programmingACM SIGPLAN Fortran Forum, 1998
- Shared Memory Programming in Metacomputing Environments: The Global Array ApproachThe Journal of Supercomputing, 1997
- Algorithms for scalable synchronization on shared-memory multiprocessorsACM Transactions on Computer Systems, 1991