ABSTRACT
The ENSEMBLE communication library exploits overlapping of message aggregation (computation) and DMA transfers (communication) for embedded multi-processor systems. In contrast to traditional communication libraries, ENSEMBLE operates on n-dimensional data descriptors that can be used to specify often-occurring data access patterns in n-dimentional arrays. This allows ENSEMBLE to setup a three-stage pack-transfer-unpack pipeline, effectively overlapping message aggregation and DMA transfers. ENSEMBLE is used to support Spar/Java, a Java-based language with SPMD annotations. Measurements on a TriMedia-based multi-processor system show that ENSEMBLE increases performance up to 39% for peer-to-peer communication, and up to 34% for all-to-all communication.
- 1.AMD Athlon TM Processor Technical Brief, Dec. 1999.Google Scholar
- 2.H. E. Bal, R. Bhoedjang, R. Hofman, C. Jacobs, K. Langendoen, T. R. uhl, and M. F. Kaashoek. Performance evaluation of the Orca shared-object system. ACM Transactions on Computer Systems, 16(1):1-40, Feb. 1998. Google ScholarDigital Library
- 3.M. Blumrich, C. Dubnicki, E. Felten, and K. Li. Protected, user-level DMA for the SHRIMP network interface. In HPCA-2, pages 154-165, San Jose, CA, Feb. 1996. Google ScholarDigital Library
- 4.S. Cadot, K. Langendoen, H. Sips, and C. v. Reeuwijk. Implementation of H-PAM, ENSEMBLE: A communication layer for the embedded heterogeneous multi-processor target of the Spar compiler. JOSES deliverable 5.2.2/1, DUT-8105-ENSEMBLE, Oct. 2000.Google Scholar
- 5.K. Chandy and C. Kesselman. CC++: A declarative concurrent object oriented programming notation. In Research Directions in Concurrent Object-Oriented Programming, pages 281-313. MIT press, 1993. Google ScholarDigital Library
- 6.High Performance Fortran Forum. High Performance Fortran Language Specification, 2.0 edition, Feb. 1997.Google Scholar
- 7.K. Lahiri, G. Lakshminarayana, A. Raghunathan, and S. Dey. Communication architecture tuners: A methodology for the design of high performance communication architectures. In 37th Design Automation Conference (DAC2000), Los Angeles, CA, June 2000. Google ScholarDigital Library
- 8.M. Lauria, S. Pakin, and A. Chien. Efficient layering for high speedcommunication: Fast Messages 2.x. In 7th High Perf. Distributed Computing Conf. (HPDC7), Chicago, Illinois, July 1998. Google ScholarDigital Library
- 9.E. Markatos and M. Katevenis. User-level DMA without operating system kernel modification. In HPCA-3, pages 322- 331, San Antonio, TX, Feb. 1997. Google Scholar
- 10.Message Passing Interface Forum. MPI: A Message-Passing Interface Standard, final report v1.0 edition, Apr. 1994.Google Scholar
- 11.S. Mukherjee, S. Sharmann, M. Hill, J. .Larus, A. Rogers, and J. Saltz. Efficient support for irregular applications on distributed memory machines. In PPoPP 95, pages 68-79, Santa Barbara, CA, July 1995. Google ScholarDigital Library
- 12.Philips Electronics, TriMedia Product Group. TM1000 Preliminary Data Book, 1997.Google Scholar
- 13.C. v. Reeuwijk, W. Denissen, H. Sips, and E. Paalvast. An implementation framework for HPF distributed arrays on message-passing parallel computer systems. IEEE Transactions on Parallel and Distributed Systems, 7(9):897-914, Sept. 1996. Google ScholarDigital Library
- 14.C. v. Reeuwijk, F. Kuijlman, and H. Sips. Extending Java with constructs for scientific computation. PDS Technical Report PDS-2001-001, Delft University of Technology, Feb. 2001. www.pds.twi.tudelft.nl/reports/2001/PDS-2001-001, accepted for publication at the Joint ACM Java Grande/ISCOPE 2001 Conference. Google ScholarDigital Library
- 15.Spar/Java compiler website. http://www.pds.twi.tudelft.nl/timber .Google Scholar
- 16.V. Sunderam. PVM: A framework for parallel distributed computing. Concurrency: Practice and Experience, 2(4):315-339, Dec. 1990. Google ScholarDigital Library
Index Terms
- ENSEMBLE: A Communication Layer for Embedded Multi-Processor Systems
Recommendations
ENSEMBLE: A Communication Layer for Embedded Multi-Processor Systems
OM '01: Proceedings of the 2001 ACM SIGPLAN workshop on Optimization of middleware and distributed systemsThe ENSEMBLE communication library exploits overlapping of message aggregation (computation) and DMA transfers (communication) for embedded multi-processor systems. In contrast to traditional communication libraries, ENSEMBLE operates on n-dimensional ...
ENSEMBLE: A Communication Layer for Embedded Multi-Processor Systems
The ENSEMBLE communication library exploits overlapping of message aggregation (computation) and DMA transfers (communication) for embedded multi-processor systems. In contrast to traditional communication libraries, ENSEMBLE operates on n-dimensional ...
Popular ensemble methods: an empirical study
An ensemble consists of a set of individually trained classifiers (such as neural networks or decision trees) whose predictions are combined when classifying novel instances. Previous research has shown that an ensemble is often more accurate than any ...
Comments