Network Subsystems in MPPs: Where Did All the Performance Go?
Resumo
Neste artigo apresenta-se resultados sobre a identificação dos principais gargalos em subsistemas de rede de processadores massivamente paralelos, abordando particularmente o Intel Paragon. Demonstra-se como técnicas de implementação correntes impedem que aplicações que utilizem protocolos como TCP / IP entre supercomputadores interligados por redes de alta velocidade alcancem taxas de transferência próximas à capacidade do meio de transmissão. Algumas possíveis soluções para o problema são sugeridas, e discute-se o que pode ser esperado de sistemas futuros. Em particular, resultados de avaliação dos benefícios de algumas técnicas são apresentados, destacando-se uma implementação de protocolos no espaço da aplicação que apresenta performance e capacidade de expansão para multiplas conexões superior aos sistemas atuais.
Referências
Joseph S. Barrera. A fast Mach network IPC implementation. In Proceedings of the Usenix Mach Symposium, 2560 Ninth Street, Suite 215, Berkeley CA 94710, November 1991. Usenix Association.
Douglas E. Comer. lnternetworking with TCP/IP, volume 1. Prentice Hall, 3rd edition, 1995.
Intel Corporation. Intel TeraFLOPs supercomputer project home page. http://www.ssd.intel.com/.
Thomas H. Dunigan. Early experiences and performance of the Intel Paragon. Technical Report ORNL/TM-12194, Oak Ridge National Laboratory, oct 1994.
Thomas H. Dunigan. Performance of ATM/OC-12 on the Intel Paragon. Technical Report ORNL/TM-13239, Oak Ridge National Laboratory, may 1996.
Alessandro Forin, David Golub, and Brian Bershad. An I/O system for Mach 3.0. In Proceedings of the Usenix Mach Symposium, 2560 Ninth Street, Suite 215, Berkeley CA 94710, November 1991. Usenix Association.
Jerry Gerner. Input/output on the IBM SP2-an overview, 1995. Available at [link].
W. Gropp, E. Lusk, anel A. Skjellum. Using MPI: Portable Parallel Programming with the Message Passing lnterface. MIT Press, 1995.
Dorgival O. Guedes and Larry L. Peterson. Eliminating the network subsystem bottleneck in MPPs. To appear, jun 1997.
Paragon XP/S product overview. Intel Corporation, 1991.
Intel Corperation. Paragon User's Cuide, oct 1993.
Jochen Liedtke. On micro-kernel construction. In Proceedings of the Fifteenth ACM Symposium on Operating System Principies. ACM, December 1995.
John LoVerso and Paul Roy. The network architecture of OSF/1 AD version 2. In OSF/RI Operating Systems Collected Papers Vol. 3. OSF Research Institute, February 1994.
Chris Maeda anel Brian N. Bershad. Networking performance for microkernels. In Proceedings of the Third Workshop on Workstation Operating Systems, May 1992.
Chris Maeda and Brian N. Bershad. Protocol service decomposition for high-performance network. In Proceedings of the Fourteenth ACM Symposium on Operating System Principles, December 1993.
S. W. O'Malley and L. L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110-143, May 1992.
Richard Rashid, Robert Baron, Alessandro Forin, David Golub, Michael Jones, Daniel Julin, Douglas Orr, and Richard Sanzi. Mach: A foundation for open systems. In Proceedings of the Second Workshop on Workstalion Operating Systems(WWOS2), sep 1989.
Paul Roy, David Black, Paulo Guedes, John LoVerso, Durriya Netterwala, Faramarz Rabii, Michael Barnett, Bradford Kemp, Michael Leibensperger, Chris Peak, and Roman Zajcew. An OSF/1 unix for massively parallel multicomputers. In OSF/Rl Operating Systems Collected Papers Vol. 2. OSF Research Institute, Cambridge, MA, October 1993.
J. Renwick. IP over HIPPI. Request for Comments (Experimental) RFC 2067, Internet Engineering Task Force, January 1997.
Subhash Saini and Horst O. Simon. Applications performance under OSF/1 AD and SUNMOS on Intel Paragon XP/2-15. In Proceedings of Supercomputing' 94, Washington, DC, nov 1994.
Peter A. Steenkiste. A systematic approach to host interface design for high-speed networks. Computer, 27(3):47-58, March 1994.
Vaidy Sunderam. Concurrent computing with PVM. In Proceedings of the Workshohp on Clusler Computing, Tallahassee, FL, December 1992. Supercomputing Computations Research lnstitute, Florida State University. Proceedings available via anonymous ftp from ftp.scri.fsu.edu in directory pub/parallel-works hop.92.
John Michael Tracey and Arindam Banerji. Device driver issues in high-performance networking. In Proceedings of the 1994 USENIX Symposium on High-Speed Networking, August 1994.
Don Tolmie and John Renwick. Hippi: Simplicity yields success. IEEE Network, January 1993.
Ronald J. Vetter. Atm concepts, architectures and protocols. Communications of the ACM, 38(2), February 1995.
Masanobu Yuhara, Brian N. Bershad, Chris Maeda, and J. Eliot B. Moss. Efficient packet demultiplexing for multiple endpoints and large messages. In USENIX Conference Proceedings, pages 153-165, San Francisco, CA, Winter 1994. USENIX.
Douglas E. Comer. lnternetworking with TCP/IP, volume 1. Prentice Hall, 3rd edition, 1995.
Intel Corporation. Intel TeraFLOPs supercomputer project home page. http://www.ssd.intel.com/.
Thomas H. Dunigan. Early experiences and performance of the Intel Paragon. Technical Report ORNL/TM-12194, Oak Ridge National Laboratory, oct 1994.
Thomas H. Dunigan. Performance of ATM/OC-12 on the Intel Paragon. Technical Report ORNL/TM-13239, Oak Ridge National Laboratory, may 1996.
Alessandro Forin, David Golub, and Brian Bershad. An I/O system for Mach 3.0. In Proceedings of the Usenix Mach Symposium, 2560 Ninth Street, Suite 215, Berkeley CA 94710, November 1991. Usenix Association.
Jerry Gerner. Input/output on the IBM SP2-an overview, 1995. Available at [link].
W. Gropp, E. Lusk, anel A. Skjellum. Using MPI: Portable Parallel Programming with the Message Passing lnterface. MIT Press, 1995.
Dorgival O. Guedes and Larry L. Peterson. Eliminating the network subsystem bottleneck in MPPs. To appear, jun 1997.
Paragon XP/S product overview. Intel Corporation, 1991.
Intel Corperation. Paragon User's Cuide, oct 1993.
Jochen Liedtke. On micro-kernel construction. In Proceedings of the Fifteenth ACM Symposium on Operating System Principies. ACM, December 1995.
John LoVerso and Paul Roy. The network architecture of OSF/1 AD version 2. In OSF/RI Operating Systems Collected Papers Vol. 3. OSF Research Institute, February 1994.
Chris Maeda anel Brian N. Bershad. Networking performance for microkernels. In Proceedings of the Third Workshop on Workstation Operating Systems, May 1992.
Chris Maeda and Brian N. Bershad. Protocol service decomposition for high-performance network. In Proceedings of the Fourteenth ACM Symposium on Operating System Principles, December 1993.
S. W. O'Malley and L. L. Peterson. A dynamic network architecture. ACM Transactions on Computer Systems, 10(2):110-143, May 1992.
Richard Rashid, Robert Baron, Alessandro Forin, David Golub, Michael Jones, Daniel Julin, Douglas Orr, and Richard Sanzi. Mach: A foundation for open systems. In Proceedings of the Second Workshop on Workstalion Operating Systems(WWOS2), sep 1989.
Paul Roy, David Black, Paulo Guedes, John LoVerso, Durriya Netterwala, Faramarz Rabii, Michael Barnett, Bradford Kemp, Michael Leibensperger, Chris Peak, and Roman Zajcew. An OSF/1 unix for massively parallel multicomputers. In OSF/Rl Operating Systems Collected Papers Vol. 2. OSF Research Institute, Cambridge, MA, October 1993.
J. Renwick. IP over HIPPI. Request for Comments (Experimental) RFC 2067, Internet Engineering Task Force, January 1997.
Subhash Saini and Horst O. Simon. Applications performance under OSF/1 AD and SUNMOS on Intel Paragon XP/2-15. In Proceedings of Supercomputing' 94, Washington, DC, nov 1994.
Peter A. Steenkiste. A systematic approach to host interface design for high-speed networks. Computer, 27(3):47-58, March 1994.
Vaidy Sunderam. Concurrent computing with PVM. In Proceedings of the Workshohp on Clusler Computing, Tallahassee, FL, December 1992. Supercomputing Computations Research lnstitute, Florida State University. Proceedings available via anonymous ftp from ftp.scri.fsu.edu in directory pub/parallel-works hop.92.
John Michael Tracey and Arindam Banerji. Device driver issues in high-performance networking. In Proceedings of the 1994 USENIX Symposium on High-Speed Networking, August 1994.
Don Tolmie and John Renwick. Hippi: Simplicity yields success. IEEE Network, January 1993.
Ronald J. Vetter. Atm concepts, architectures and protocols. Communications of the ACM, 38(2), February 1995.
Masanobu Yuhara, Brian N. Bershad, Chris Maeda, and J. Eliot B. Moss. Efficient packet demultiplexing for multiple endpoints and large messages. In USENIX Conference Proceedings, pages 153-165, San Francisco, CA, Winter 1994. USENIX.
Publicado
07/10/1997
Como Citar
GUEDES, Dorgival O.; PETERSON, Larry L..
Network Subsystems in MPPs: Where Did All the Performance Go?. In: INTERNATIONAL SYMPOSIUM ON COMPUTER ARCHITECTURE AND HIGH PERFORMANCE COMPUTING (SBAC-PAD), 9. , 1997, Campos do Jordão/SP.
Anais [...].
Porto Alegre: Sociedade Brasileira de Computação,
1997
.
p. 333-348.
DOI: https://doi.org/10.5753/sbac-pad.1997.22634.
