Nanette J. Boden, Danny Cohen, Robert E. Felderman, Alan E. Kulawik, Charles L. Seitz, Jakov N. Seizovic, and Wen-King Su. Myrinet|a gigabit-per-second local-area network. IEEE Micro, 15(1):29-38, February 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....that expose the vulnerability of such microprocessor based network systems to faults. In Section 5, we discuss some implementation details and report on the performance and overhead of the fault tolerance scheme. We conclude the paper in Section 6. 2 Myrinet: An Example System Myrinet [3] is a cost e ective, high bandwidth (2 Gb s) packet communication and switching technology from Myricom Inc [11] It employs wormhole switching, backpressure ow control and source routing to achieve low latency ( 8 s) transfer of messages. A Myrinet network consists of point to point, ....

N. J. Boden, D. Cohen, R. E. Felderman, A. E. Kulawik, C. L. Seitz, J. N. Seizovic, and W.-K. K. Su. Myrinet | A gigabit-per-second local-area-network. IEEE Micro, 15(1):29-36, Feb 1995.

....is managed by the operating system, and so performance is less good than for native implementations, but this approach is cheap and highly exible. The third approach is to use an intelligent NIC. Intel s proof of concept[3] implementation and Berkeley VIA[4] both make use of Myricom s Myrinet[5] a gigabit class, programmable, user M VIA also supports custom VIA hardware. level accessible NIC. As with custom hardware implementations, scalability is limited by the resources on the NIC. Implementations that use specialised hardware can certainly achieve signi cantly higher ....

Nanette Boden, Danny Cohen, Robert Felderman, Alan Kulawik, Charles Seitz, Javoc Seizovic, and Wen-King Su. Myrinet | A Gigabit-per-Second Local-Area Network. IEEE Micro, 15(1), 1995.

....Keywords Networks of workstations, irregular topologies, up down routing, deadlock avoidance, genetic algorithms. This work was partly supported by the Spanish CICYT under Grant TIC97 0897 C04, and Caja Castilla La Mancha 1 Introduction High speed NOWs, like Autonet [15] Myrinet [1] and Servernet [9] provide an inexpensive alternative for building parallel computers with high communication bandwidth and reduced latency. These networks usually consist of a set of switches connected by point to point links, and of hosts linked to switches through a network interface card ....

....parallel computers with high communication bandwidth and reduced latency. These networks usually consist of a set of switches connected by point to point links, and of hosts linked to switches through a network interface card (NIC) Up down routing is broadly used for routing in high speed NOWs [15, 1]. However, it is well known that the performance of up down routing is strongly dependent on the network topology [11] The generation of up down routing tables requires the construction of a directed graph by assigning directions to the operational links in the network. As we will see later, ....

N.J. Boden, D. Cohen, and R.E. Felderman, \Myrinet { a gigabit per second local area network", IEEE Micro, pages 29-36, February 1995.

....SHUFFLENET A. Wormhole Routing Networks Wormhole routing networks with backpressure flow control require bidirectional links, thereby being a natural context for the use of the bidirectional shufflenet. Wormhole routing is often the scheme of choice in low latency high speed networks [14] 21] [3], 16] In wormhole routing, the (variable size) unit of information transfer is a worm, which can range in length from a few bytes to several thousand bytes. Low latency is achieved by using cut through or wormhole routing instead of the conventional store and forward approach: each intermediate ....

....for the entire worm to be assembled. Thus, the worm can stretch across several nodes and links at any one time. The key feature of wormhole routing high speed networks is that they can be operated in an asynchronous and unslotted fashion. An example of a wormhole routing network is Myrinet [20] [3]. Typically, wormhole routing networks are based on very simple network protocols and therefore are prone to congestion. Congestion is a crucial problem at very high speeds, where large amounts of information can be lost when network resources become unavailable. To prevent congestion some form ....

N. Boden, D. Cohen, R. Felderman, A. Kulawik, C. L. Seitz, J. Seizovic, and W.-K. S. Myrinet, "A gigabit-per-second local-area network," IEEE Micro, vol. 15, pp. 29--36, Feb. 1995.

.... computationally intensive scientific applications [12] While much of the scientific community still operates in the domain of supercomputers and multi processor computers, high speed networks of workstations are fast becoming a popular alternative [12] Emerging network technologies like Myrinet [5], ATM, and SCI [21] enable data transfer at the rates of Gigabits per second, thus achieving speeds comparable to that between processors in a single parallel computer. Low level messaging layers like BDM [11] Fast Messages [17] BIP [4] and GM [10] are designed to provide highly efficient ....

Boden, J. Nanette, D. Cohen, R. E. Felderman, A. E. Kulawik, C. L. Seitz, J. N. Scizovic, and W. K. Su. A Gigabit-per-Second Local Area Network. IEEE-Micro, 15 (1): 29-36. February 1995.

....may be posted at one time. I O Bus Network Processor Host Memory Host Chip Host DMA Processor Network DMA LANai SRAM The targeted platform is the LHPC PoPC which is a cluster of Pentium Pro 200 Mhz running Linux. Each workstation is equipped with a Myricom PCI network interface card [2] on the PCI Bus. The gure shown here shows the details of the interface card. It contains a host DMA engine, which moves data from host main memory to the SRAM on the network interface, a network DMA engine, which moves the data from the SRAM into the network, and a LANai processor, which ....

Nanette J. Boden, Danny Cohen, Robert E. Felderman, Alan E. Kulawik, Charles L. Seitz, Jakov N. Seizovic, and Wen-King Su. Myrinet  A Gigabit-per-Second LocalArea Network. IEEE MICRO, pages 2936, February 1995. http://www.myri.com/ research/publications/index.html.

.... to communication performance [43] ii) high connectivity is super uous because processes of a parallel program only needed to communicate with their peers running on the same system, and (iii) parallel programs and machines were carefully designed so that messages are lost only very rarely [7, 22]; a message loss is typically treated as a catastrophic event either the program or 1 the system will crash and need to be restarted. The need for a transport layer is thus eliminated by the programmer s fault model and the application requirements. For the emerging class of large scale ....

Boden, N. J., Cohen, D., Felderman, R. E., Kulawik, A. E., Seitz, C. L., Seizovic, J. N., and Su, W.-K. Myrinet|A Gigabit-per-Second Local-Area Network. IEEE Micro 15, 1 (Feb. 1995), 29-38.

....University ecap,anand,jothi,manoj,tony hpcl.cs.msstate.edu Abstract Clock synchronization is a fundamental requirement for any real time distributed system operating with global schedules. This paper describes the design and implementation of a high accuracy (4 s) global clock on a Myrinet [4] gigabit s system area network of PCs with considerably low software overheads. The global clock is based on a master slave internal clock synchronization scheme [15] A novel approach has been adopted to improve synchronization accuracy. The programmability of the Myrinet interface card and the ....

....with considerably low software overheads. The global clock is based on a master slave internal clock synchronization scheme [15] A novel approach has been adopted to improve synchronization accuracy. The programmability of the Myrinet interface card and the presence of an on board Real Time Clock [4] have been utilized to counter the undesirable effects of unpredictability in the latency of clock messages. The resulting synchronization facilitates global scheduling of distributed real time tasks, and provides a framework to build support for Quality of Service in distributed high performance ....

[Article contains additional citation context not shown here]

N.J.Boden, D. Cohen, R. E. Felderman, A. E. Kulawik, C. L. Seitz, J. N. Seizovic, and W. K. Su. "A Gigabit-per-second Local Area Network", IEEE Micro15, pp 29-36, Feb 1995.

....We describe some of the lessons learned from our experience with programming Myrinet networks, as well as feedback to NIC developers from a user s perspective. Keywords Gigabit networking, Myrinet, communication protocols, performance, reliability, portability Myrinet Sidebar Myrinet [1], 12] is a gigabit per second network with full duplex 1.28 1.28 gigabit per second links, switch ports, and interface ports. The host interface boards contain a custom processor, the LANai, that can execute a control program in order to handle direct interaction with host processes; sending, ....

Nanette J. Boden, Danny Cohen, Robert E. Felderman, Alan E. Kulawik, Charles L. Seitz, Jakov N. Seizovic, and Wen-King Su. A Gigabit-per-Second Local-Area Network. IEEE-Micro, 15(1):29--36, February 1995.

....scheduler. This paper documents the BDT library as it is now BDT is a work in progress and will likely change. 1 Introduction The BDT (BullDog 1 Thread) library consists of a set of library routines for building multi threaded Myrinet Control Programs (MCPs) for the Myricom LANai processor[1, 4]. Initially, the motivation for BDT was to support multiple context debugging with our MCP debugger, BDMD[2] and for experimenting with the scheduling of real time communications on a communication co processor. 2 Using the BDT Library 2.1 Installing BDT Installing BDT is relatively simple. You ....

Nanette J. Boden, Danny Cohen, Robert E. Felderman, Alan E. Kulawik, Charles L. Seitz, Jakov N. Seizovic, and Wen-King Su. A Gigabit-per-Second Local-Area Network. IEEE-Micro, 15(1):29--36, February 1995.

....and active networking experiments. Robust reliability and low processor overhead distinguish this effort from others. 1 Introduction The BDM (BullDog 1 MCP) software consists of a set of host library routines and a Myrinet Control Program (MCP) for the Myricom LANai network processor [1, 15]. BDM was designed to provide a capable yet simple MCP useful for experiments with new protocols and different levels of service in support of systems such as MPI [7] and Packetway [8] Also, we want to provide the lowest latency and highest bandwidth possible consistent with robust reliability. ....

Nanette J. Boden, Danny Cohen, Robert E. Felderman, Alan E. Kulawik, Charles L. Seitz, Jakov N. Seizovic, and Wen-King Su. A Gigabit-per-Second Local-Area Network. IEEE-Micro, 15(1):29--36, February 1995.

No context found.

Nanette J. Boden, Danny Cohen, Robert E. Felderman, Alan E. Kulawik, Charles L. Seitz, Jakov N. Seizovic, and Wen-King Su. Myrinet|a gigabit-per-second local-area network. IEEE Micro, 15(1):29-38, February 1995.

No context found.

Boden, N.J., et al., {Myrinet}---A Gigabit-per-Second Local-Area Network. IEEE Micro, 1995. 15(1): p. 29--36.

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