T. M. Anderson and R. S. Cornelius. High- performance switching with Fibre Channel. In Digest of Papers Compcon 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....peak bandwidth of about 300 MB s, the IBM SP2 of about 35 ps and 100 MB s respectively, whereas the typical values for a classical Ethernet interconnected cluster are 1 ms and 1.2 MB s respectively. The new high speed Local Area Networks (LANs) available today (ATM [4] FDDI [10] Fibrechannel [1], Myrinet [2] offer comparable hardware latency and bandwidth to the proprietary interconnect found on MPPs. The introduction of these enabling technologies shifts the focus of the MPP versus cluster comparisons from performance more general considerations of system scalability, reliability, ....

T. M. Anderson and R. S. Cornelius. High- performance switching with Fibre Channel. In Digest of Papers Compcon 1992.

....of advantages over their major competitors (massively parallel processors based on workstation processors) These advantages can include lower cost, a larger software base, and greater accessibility. Further, the advent of high performance network interconnects such as ATM [7] Fibre Channel [4], FDDI [13] and Myrinet [6] present the possibility that workstation clusters can deliver good performance on a broader range of parallel computations. Achieving efficient communication is the major challenge in synthesizing effective parallel machines from networks of workstations. ....

.... [27] atop UDP or TCP [28] is perhaps the most popular) The communication primitives in these libraries have typically exploited operating system communication services, running atop 10 Mb s Ethernet, or more recently some higher speed physical media such as FDDI [13] ATM [7] or Fibre Channel [4]. While such facilities are useful for coarse grain decoupled parallelism, they suffer from high software communication overhead (operating system calls) and low achieved bandwidth (media limits or software overhead) and thus cannot support more tightly coupled or finer grained parallelism. ....

T. M. Anderson and R. S. Cornelius. High-performance switching with Fibre Channel. In Digest of Papers Compcon 1992, pages 261--268. IEEE Computer Society Press, 1992. Los Alamitos, Calif.

....and peak bandwidth of about 300 MB s, the IBM SP2 of about 30 s and 40 MB s respectively, whereas the typical values for a classical Ethernet interconnected cluster are 1 ms and 1.2 MB s respectively. The new high speed Local Area Networks (LANs) available today (ATM [4] FDDI [11] Fibrechannel [1], Myrinet [2] offer comparable hardware latency and bandwidth to the proprietary interconnect found on MPPs. The introduction of these enabling technologies shifts the focus of the MPP versus cluster comparisons from performance more general considerations of system scalability, reliability, ....

T. M. Anderson and R. S. Cornelius. High-performance switching with Fibre Channel. In Digest of Papers Compcon 1992, pages 261--268. IEEE Computer Society Press, 1992. Los Alamitos, Calif.

....compete with supercomputers in aggregate performance. Many supercomputers today are microprocessor based [35, 34, 24, 12, 37] because such microprocessors deliver instruction and floating point processing rates in excess of 1 billion operations per second. In addition, emerging high speed networks [17, 3, 6, 19, 25]) provide the aggregate capability for supercomputing performance. While the development of these technologies has long been anticipated, their potential impact has been accentuated by the development of new communication technology described below. Over the past four years, the research ....

....approach. By embracing the use of legacy protocols with their inherent latency bandwidth limitations, the Beowulf approach is better suited to large grained applications. The Beowulf concept emphasizes performance price ratio over generality of use. The advent of fast interconnect technologies [17, 3, 6, 19, 25]) has changed the role of cluster architectures from machines for coarse granularity applications to inexpensive general purpose parallel computers. Over the past four years, the research community has produced dramatic progress in delivering hardware communication performance to applications ....

T. M. Anderson and R. S. Cornelius. High-performance switching with Fibre Channel. In Digest of Papers Compcon 1992, pages 261--268. IEEE Computer Society Press, 1992. Los Alamitos, Calif.

....computing systems increasingly desirable. The Beowulf cluster approach [3] which relies on the pervasive Fast Ethernet technology (and essentially zero incremental cost) produces clusters that support coarse grained, loosely coupled parallelism. The faster system area network (SAN) technologies [1,2,4,10,13] Performance Enhancements for HPVM 3 provide hardware enablers for tightly coupled clusters. Significant advances from the research community in the past five years have made it possible to deliver hardware communication performance to the applications Fast Messages (FM) 17] Active Messages (AM) ....

T. M. Anderson and R. S. Cornelius. High-performance switching with fibre channel. In Digest of Papers Compcon 1992, pages 261--268. IEEE Computer Society Press, 1992. 2

....of workstation a truly general purpose machine. What is making workstation clusters more attractive than before is the introduction of new communication technologies, with much improved performance with respect to the current networks. The fast LANs available today (ATM [6] FDDI [9] Fibrechannel [2], Myrinet [4] in terms of latency and bandwidth can be comparable to the proprietary interconnection networks found in MPPs. 1.3 The Approach The simple substitution of new, fast network hardware for the old, without a corresponding change to the existing communication software represents only ....

.... numsets, int numreps, int datasize, int verbose) numsets = 5; numreps = 20; datasize = 16; verbose = 1; if (argc 1) numsets = atoi(argv[1] if ( numsets maxnumsets) printf( Too many sets ( d vs d) n , numsets, maxnumsets) exit( 1) if (argc 2) numreps = atoi(argv[2]) if ( numreps maxnumreps) printf( Too many reps ( d vs d) n , numreps, maxnumreps) exit( 1) if (argc 3) datasize = atoi(argv[3] if ( datasize maxdatasize) printf( Too much data ( d vs d) n , datasize, maxdatasize) exit( 1) 48 if (argc 4) verbose = ....

[Article contains additional citation context not shown here]

T. M. Anderson and R. S. Cornelius. High-performance switching with Fibre Channel. In Digest of Papers Compcon 1992, pages 261--268. IEEE Computer Society Press, 1992. Los Alamitos, Calif.

....and peak bandwidth of about 300 MB s, the IBM SP2 of about 30 s and 40 MB s respectively, whereas the typical values for a classical Ethernet interconnected cluster are 1 ms and 1.2 MB s respectively. The new high speed Local Area Networks (LANs) available today (ATM [4] FDDI [11] Fibrechannel [1], Myrinet [2] offer comparable hardware latency and bandwidth to the proprietary interconnect found on MPPs. The introduction of these enabling technologies shifts the focus of the MPP versus cluster comparisons from performance more general considerations of system scalability, reliability, ....

T. M. Anderson and R. S. Cornelius. High-performance switching with Fibre Channel. In Digest of Papers Compcon 1992, pages 261--268. IEEE Computer Society Press, 1992. Los Alamitos, Calif.

....y Department of Computer Science, University of Illinois at Urbana Champaign, 1304 W. Springfield Ave. Urbana, IL 61801, USA z Dipartimento di Informatica e Sistemistica, Universit a di Napoli Federico II , via Claudio 21, 80125 Napoli, Italy interconnects such as ATM [7] Fibre Channel [4], FDDI [13] and Myrinet [6] present the possibility that workstation clusters can deliver good performance on a broader range of parallel computations. Achieving efficient communication is the major challenge in synthesizing effective parallel machines from networks of workstations. ....

.... [27] atop UDP or TCP [28] is perhaps the most popular) The communication primitives in these libraries have typically exploited operating system communication services, running atop 10 Mb s Ethernet, or more recently some higher speed physical media such as FDDI [13] ATM [7] or Fibre Channel [4]. While such facilities are useful for coarse grain decoupled parallelism, they suffer from high software communication overhead (operating system calls) and low achieved bandwidth (media limits or software overhead) and thus cannot support more tightly coupled or finer grained parallelism. ....

T. M. Anderson and R. S. Cornelius. High-performance switching with Fibre Channel. In Digest of Papers Compcon 1992, pages 261--268. IEEE Computer Society Press, 1992. Los Alamitos, Calif.

....hundreds of thousands instructions. Workstation clusters have recently become attractive for high performance computation due to the introduction of new communication technologies with much improved performance. The fast Local Area Networks (LANs) available today (ATM [5] FDDI [8] Fibrechannel [1], Myrinet [3] are, in terms of latency and bandwidth, comparable to the proprietary interconnect found on MPPs. However, without a radical change in the way communication protocols are implemented, applications will not be able to reap the benefits of this new technology. Existing protocol ....

T. M. Anderson and R. S. Cornelius. High-performance switching with Fibre Channel. In Digest of Papers Compcon 1992, pages 261--268. IEEE Computer Society Press, 1992. Los Alamitos, Calif.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC