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, February 1995, pp. 29 - 36.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....processor, and not all platforms implement pre emptive scheduling. parallel computing nor does it address more involved or irregular communication needs. Currently several projects are under way that achieve increased bandwidth, reduced latency, and better reliability within workstation clusters [1, 2, 21]. These projects will fill the gaps between desktop SMPs, parallel computers, and Internet based meta computing. A similar gap can be noticed on the software side. With improved cluster interconnection technology and growing cluster reliability and availability, Java s current mechanisms become ....

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

....some of previous work on improving networking performance using programmable network interfaces. Network interfaces based on the Tigon and the Myricom LANai adapters are two widely used programmable network interfaces. The Myricom LANai adpaters are for a system area network called Myrinet [6], whereas the Tigon is an Ethernet interface controller as described in Section 3.1. The projects that used these programmable network interfaces range from implementing various optimizations for basic send and receive processing to implementing all or parts of network protocol processing on the ....

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--36, 1995.

....but also taps into the aggregate bandwidth of these disks to deliver a highperformance and scalable storage service. PVFS [3] is one example of such file systems and it can achieve multiple GBytes sec I O throughputs [4] without any additional cost if the cluster is connected through Myrinet [5] or Gigabit Ethernet. However, like disk arrays [6] without any fault tolerance, these parallel storage systems are too unreliable to be useful since the failure rate of a cluster node, compounded by the failures of cluster hardware components, including CPU, disk, memory and network, and the ....

Nanette J. Boden, Danny Cohen, Robert E. Felderman, Alan E. Kulawik, Charles L Seitz, Jakov N. Seizovic, and Wen-King Su, "Myrinet: A gigabitper -second local area network," IEEE Micro, 15(1): 29-36, 1995

....by System Load. 4 Related Work Communication CoProcessors. A number of NI based research projects have focused on providing low latency message passing over cluster interconnects like ATM, Myrinet, FDDI and HIPPI[15, 40, 58, 57] by using intel ligent NIs equipped with programmable CoProcessors [10, 21, 38, 45, 49]. In part, such work is motivated by the presence of programmable CoProcessors in NIs. More importantly, past research has demonstrated improved performance derived from NI reprogramming for control operations that touch multiple machines like collective communications[53] transactions[48] ....

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, Feb. 1995.

....module (8 port 1000Base SX) At the price performance point of 128 nodes, the cost of this network is 20 of the cluster while the E7 solution is 13 . High Performance Network As a representative of the expensive networks we are considering Myrinet 2000. A technical introduction is given in [3]. The important difference to the previous networking concepts is the emphasis on expensive network interfaces and low cost high speed switches. At the price performance point of 128 nodes the overall cost of Myrinet is about 35 of the total cost of the cluster. In our cluster a 32 dual processor ....

Nanette J. Boden, 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--36, February 1995.

....are being used to couple nodes in clusters, storage, and even to supplant local area networks. This study is performed in the context of Fast Messages [14] a user level interface designed to deliver the performance of high performance networks to applications. FM 2. 1 exploits Myrinet hardware [3] to achieve one way latencies of 10s and peak bandwidths of over 90 MB s. This level of performance is comparable to local I O bus performance, and is similar to emerging all hardware implementations of the Virtual Interface architecture [20] The Myrinet is a high speed SAN that provides ....

....of the transport only. The ping pong tests were run on two machines. Each machine included dual 300 MHz Pentium II processors (with 512 kB of cache each) and 512 MB of main memory running Windows NT Server 4.0 SP3. The machines are connected by shared 100Mbit Ethernet and 160MB s Myrinet switches [3]. To reduce transient interference, the ping pong tests include many round trips, and take the minimum of two sets of runs. The one way latency of the FM loadable transport tracks the latency of the underlying FM layer closely (see Figure 2) delivering most of the underlying performance. The ....

[Article contains additional citation context not shown here]

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-- 36, February 1995. Available from http://www.myri.com/research/publications/Hot.ps.

....module (8 port 1000Base SX) At the price performance point of 128 nodes, the cost of this network is 20 of the cluster while the E7 solution amounts 13 . High Performance Network As a representative of the expensive networks we are considering Myrinet 2000. A technical introduction is given in [3]. The important difference to the previous networking concepts is the emphasis on expensive network interfaces and low cost high speed switches. At the priceperformance point of 128 nodes the overall cost of Myrinet is about 35 of the total cost of the cluster. In our cluster a 32 dual processor ....

Nanette J. Boden, 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--36, February 1995.

....the ability to stripe data across heterogeneous networks, the ability to fail over from one network onto a heterogeneous network, and the ability to add data filters and other features dynamically, remotely, and even on per connection bases. I. INTRODUCTION HEN gigabit networks, such as Myrinet [1], Giganet [2] and Gigabit Ethernet [3] started to become commonplace on workstation clusters, the communication performance bottleneck shifted from network hardware to the messaging software. A number of highly optimized messaging layers Fast Messages [4] Active Messages [5] and U Net [6] to ....

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, vol. 15, no. 1, pp. 29 36, Feb. 1995, Available: http: //www .myri. com/research/publications/Hot. ps.

....machines, with thousands of nodes. The Computational Plant (Cplant) at Sandia National Laboratories includes several large scale parallel computers composed of commodity computing and networking components. In order to enhance scalability, Cplant uses a high performance interconnect, Myrinet [5], and a custom lightweight communication protocol based on Portals [6] When the run time environment of Cplant launches a job, it first identifies a group of active worker nodes, organizes them in a logical tree structure and then fans out the executable. The experimental results in [6] show that ....

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

....is managed by the operating system, and so performance is less good than for native implementations, but this approach is cheap and highly flexible. 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 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 significantly higher performance M VIA also supports custom VIA ....

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.

....project [34] at Sandia National Laboratories is the closest project in spirit to ours in that it identifies poor resource management performance as a problem worth studying and approaches the problem by replacing a traditionally nonscalable algorithm with a scalable one. Given the same Myrinet [6] based platform the STORM mechanisms would likely be implemented similarly to Cplant s. However, on a platform such as QsNET, which boasts hardware collectives, STORM is able to exploit the underlying hardware to improve job launching performance by a hundredfold. BProc [19] the Beowulf ....

Nanette J. Boden, Danny Cohen, Robert E. Felderman, Alan E. Kulawick, Charles L. Seitz, Jakov N. Seizovic, and Wen-King Su. Myrinet: A gigabit-per-second local area network. IEEE Micro, 15(1):29--36, February 1995. Available from http://www.myri.com/research/ publications/Hot.ps.

....build increasingly powerful cluster configurations for parallel computing has been enabled by the availability of increasingly capable network interconnects and software. Interconnect technologies such as Myrinet provide moderately high bandwidth and support kernel bypass messaging for low latency [1][2] These interconnects have been primarily utilized in clusters whose compute nodes contain only a few processors each [3] 4] The Berkeley CLUMPS project was one of the first to use larger and more capable SMP systems as the basic building blocks for a cluster [5] 6] SMP systems with larger ....

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, 15(1):29-36, February 1995.

....transfer bandwidth roughly in half, it provides significantly reduced latencies for small transfers by avoiding the need for prenegotiation with the receiving node. Networks of workstations with interprocessor communication performance rivaling that of the CM 5 are rapidly becoming reality [7, 56, 77, 80]. For example, Thekkath et al. 78] describe the implementation of a specialized data transfer mechanism implemented on a pair of 25 MHz DECstations connected with a first generation FORE ATM network. They report round trip times of 45 microseconds (1125 cycles) to read 40 bytes of data from a ....

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, pages 29--36, February 1995.

....and trends that support the above prospect. First, parallel machines will become ubiquitous. There is a strong economical demand that parallel intensive applications should run not only on dedicated parallel machines (e.g. CM5, AP1000, T3D, and Paragon) but also on networks of workstations [5, 33]. Recent research [2] has demonstrated that, with suitable communication infrastructures, intensive applications perform well on networks of workstations. Concurrent languages provide ease of programming, portability, and efficiency of applications on such computing environments. Second, ....

Nanette J. Boden, Danny Cohen, Robert E. Felderman, Alan E. Kulawik, Charles L. Seitz, Jakov N. Seizovic, and Wen-King Su. Myrinet: A gigabit-persecond local area network. IEEE Micro, 15(1):29--36, February 1995. 7 This paper describes algorithms and results by message passing C on CM5, and we are now preparing the result in Schematic.

....performance critical components in modern supercomputers. A cursory, and admittedly non comprehensive, look at the state of the art in networking technology in high performance computers indicates a number of notable players. The list includes Gigabit Ethernet [14] Giganet [17] SCI [6] Myrinet [1], GSN (HIPPI 6400) 16] 1 , just to name a few. These network solutions are different in terms of programmability, scalability, topology and performance. At the low end of the performance spectrum we have Gigabit Ethernet which provides a cost effective solution for system area The work was ....

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

....traffic generated by SPMD parallel applications. keywords: Interconnection Networks, Wormhole Routing, Parallel Computer Architecture, Performance Modeling Evaluation, Network Computing 2 1 Introduction With the recent commercial introduction of low cost, high performance communication routers [1] [2] featuring few hundreds of nanoseconds switch latencies, gigabytes per second of aggregate bandwidth and low transmission error rates, networks of workstations can provide substantially more performance than conventional local area networks and can be a cost effective alternative to ....

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

....contention resolution mechanisms. Keywords: Interconnection Networks, Performance Evaluation, User level Communication, Operating System Bypass. 1 Introduction Some interconnect technologies used in high performance computers include Gigabit Ethernet [21] Giganet [24] SCI [9] Myrinet [2] and GSN (HIPPI 6400) 23] 1 . Each one provides a different level of programmability, raw performance and integration with the operating system. InfiniBand [1] has been recently proposed as a standard for communication between processing nodes and I O devices as well as for interprocessor ....

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

....processor has a 256 KB L2 cache and each node contains 64 MB of main memory. The cache coherent 50 MHz MBus connects the processors and memory. I O devices reside on the 25 MHz SBus, which connects to the MBus via a bridge. All nodes run Solaris 2.4. Each node contains a Myrinet network interface [3], which consists of a 7 MIPS custom processor (LANai) and 128KB of memory. The LANai performs limited protocol processing and schedules DMA transfers between the network and LANai memory or LANai memory and SPARC memory. The 16 nodes used in this paper are connected with three Myrinet 8 port ....

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):29-36, February 1995.

....hardware have become critical components in achieving high performance. Such components will greatly impact the design, architecture, and use of the aforementioned systems in the future. Key players in high speed interconnects include Gigabit Ethernet (GigE) 11] GigaNet [13] SCI [5] Myrinet [1], and GSN (HiPPI 6400) 12] These interconnect solutions differ from one another with respect to their architecture, programmability, scalability, performance, and ease of integration into large scale systems. While GigE resides at the This work was supported by the U.S. Department of Energy ....

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

....how this model can be used for comparing the performance of different collective communication algorithms. We also show how the effect of heterogeneity on the performance of collective communication operations can be predicted. 1 Introduction The availability of modern networking technologies [1, 4] and cost effective commodity computing boxes is shifting the focus of high performance computing systems towards Networks of Workstations (NOW) The NOW systems comprise of clusters of PCs workstations connected over the Local Area Networks (LAN) and provide an at This research is supported ....

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

.... is implemented under Illinois Fast Messages (FM) a user level messaging layer developed at the University of Illinois at Urbana Champaign [15] Fast Messages is a high performance messaging layer that bypasses the operating system to provide direct access to an underlying Myricom Myrinet [1] and thereby achieve high performance. Details of FM can be found in [13, 15] We also employ an implementation of the Message Passing Interface (MPI) built atop FM, called MPI FM [11] for some of the benchmarks. Implementing Spin block in Fast Messages Fast Messages was enhanced with a ....

Nanette J. Boden, Danny Cohen, Robert E. Felderman, Alan E. Kulawik, Charles L. Seitz, Jakov N. Seizovic, and Wen-King Su. Myrinet---a gigabit-persecond local-area network. IEEE Micro, 15(1):29--36, February 1995. Available from http://www.myri.com/research/publications/Hot.ps.

....is characterized by message latency, page fault, and interrupt times that are relatively large compared with memory and network bandwidth (Table 3) As a consequence, a roundtrip communication for either a page or lock transfer is at best on the order of a millisecond. Current network technologies [6, 13, 7], as well as aggressive software for fast interrupts, exceptions [30] and virtual memory mapped communication [10, 11] have brought such latencies down signi cantly to the neighborhood of a couple of microseconds. An interesting question is to what extent our results are speci c to the Paragon ....

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):29-36, February 1995.

....than a cluster of common desktops. On the other hand, not being able to open the ePCs also means that it is not possible to modify their hardware. Hence it is not possible to add memory, nor to add a second network adapter or to upgrade the network adapter to a low latency network such as Myrinet [4], SCI [5] or VIA [6] which are typically used for interconnection in computational clusters. b. Connectivity The network technology being constrained by the hp e PC connectivity, the nodes are connected to five switches by Ethernet 100 links. The switches are hp ProCurve 4000 Switches ....

....as provided in Linux 2.2.4 kernels. This choice badly influences HPL performance numbers because of the latency overhead of the TCP stack. Latency is caused by data extra copies between user space and the network card memory. This phenomenum can only be avoid using specific hardware (Myrinet [4], SCI [5] or specific low latency protocols (GAMMA [7] or architectures (VIA [6] Until now, none of these solutions achieved the same stability, low cost and compatibility as TCP. This latency is widely introduced by HPL with a great number of medium sized messages. Several parameters have ....

Nanette J. Boden, Robert E. Felderman, Alan E. Kulawik, Charles L. Seitz, Jakov N. Seizovic, and Wen-King Su, "Myrinet - A Gigabit per Second Local Area Network," in IEEE-Micro, February 1995, vol. 15(1), pp. 29-36. 8

....a display PC. Each PC is a Dell Precision Workstation 420 with a 733Mhz Pentium III CPU, an Intel 840 chipset with 133Mhz front side bus, 256MB of dual channel RDRAM memory, and a nVidia GeForce II chip based graphics card. Each PC runs Microsoft Windows 2000. The communication network is Myrinet [3]. Each PC uses a previous generation, 32 bit 33Mhz PCI network interface card that has 2MB of SDRAM and a 33Mhz LANai 4 network processor. The 26 PCs are networked together with a 32 port switch which is implemented with eight 8 port crossbar switches. We have used the GM driver for Windows 2000 ....

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--36, February 1995.

....by 4 grid to form a seamless image over an 18 by 8.5 rear projection screen. The resolution of each projector is 1024 by 768, and thus the cumulative resolution of the entire wall is 6K by 3K (18M pixels) Each projector is driven by a commodity PC connected to a network (Ethernet and or Myrinet [3]) The system also contains PCs for tracking user input, synthesizing sounds for 14 speakers, and running applications. The total cost of the system is around 200K. As compared to traditional rendering systems, our architecture has several advantages. First, we use commodity hardware components, ....

Nanette J. Boden, Danny Cohen, Robert E. Felderman, Alan E. Kulawik, Charles L. 13 Seitz, Jakov N. Seizovic, and Wen-King Su. Myrinet: A gigabit-per-second local area network. IEEE MICRO, 15(1):29--36, February 1995.

....handling multicast traffic. The architecture of today s routers does not allow for scalability, e.g. by the addition of more processing nodes or faster bus technology. At the same time there has been an emergence of very high speed, economical and scalable SAN LAN interconnects, such as Myrinet [2] and ServerNet [1] These SAN LAN interconnects provide bandwidths of gigabits per second. For example, a Myrinet link composed of a full duplex pair of channels provides a bandwidth of 1.28 Gbps. These switches can be incrementally expanded to provide higher aggregate throughput and thus ....

Nanette J. Boden et al, "Myrinet -- A Gigabit-persecond Local-area Network", IEEE Micro, Feb 1995.

....The remote network interface card checks this table to translate network address into physical address and see if it can operate on this physical address. Remote memory operation provides Read Request, Read Response, Write Request, Write Response, and Unacknowledged Write. 5.1. 4 Myrinet Myrinet[6] is also a scalable, switching network. It s network topology is primary aimed at 2dimensional mesh, but not limited to it. Originally, its physical transmission media is parallel wire cable. Optic fiber is used in near future. It supports variable sized packet from two reasons. 1. The routing ....

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, Vol.15, No.1 pp.29--36 (February 1995)

....The output of each server is connected to a projector which projects onto a common screen. These projectors are configured into a tiled array so that their outputs produce a single large image. The client and servers are connected via a high speed network. Our current installation uses Myrinet [1], but WireGL is not bound to a specific network technology. As long as the network can support basic message passing, our system can easily be modified to support it. WireGL runs on TCP IP and shared memory in addition to Myrinet GM, and a VIA port is in progress. WireGL also makes no assumptions ....

Nanette Boden, Danny Cohen, Robert Felderman, Alan Kulawik, Charles Seitz, Jakov Seizovic, and Wen-King Su. Myrinet: A gigabit-per-second local area network. IEEE Micro, pages 29--36, February 1995.

....server to provide class files. Finally, we did not yet implement activation of remote objects on demand. 5 ParaStation Network ParaStation [24] is a communication technology for connecting off the shelf workstations to form a supercomputer. The current ParaStation hardware is based on Myrinet [2], fits into a PCI slot, employs technology used in massively parallel machines, and scales up to 4096 nodes. ParaStation s user level message passing software preserves the low latency of the communication hardware by taking the operating system out of the communication path, while still ....

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

....Fast Messages is a high performance messaging layer that uses a user level library to provide messaging primitives without the overhead of domain crossing that would be required by a kernel resident device driver. The implementation of Illinois Fast Messages we used ran on a Myrinet network [2] connecting eight SPARCstation 2 workstations. The Myrinet switch provides a relatively low latency, high bandwidth interconnection for workstations. The Myrinet 2.3 interface boards we used had a slow CISC processor operating at the 20 MHz speed of the SPARCstation s SBUS peripheral bus; however, ....

Nanette J. Boden, Danny Cohen, Robert E. Felderman, Alan E. Kulawik, Charles L. Seitz, Jakov N. Seizovic, and Wen-King Su. Myrinet---a gigabit-persecond local-area network. IEEE Micro, 15(1):29--36, February 1995. Available from http://www.myri.com/research/publications/Hot.ps.

....performance critical components in modern supercomputers. A cursory, and admittedly non comprehensive, look at the state of the art in networking technology in high performance computers indicates a number of notable players. The list includes Gigabit Ethernet [14] Giganet [17] SCI [6] Myrinet [1], GSN (HIPPI 6400) 16] 1 , just to name a few. These network solutions are di erent in terms of programmability, scalability, topology and performance. At the low end of the performance spectrum we have Gigabit Ethernet which provides a coste ective solution for system area networks. Giganet, ....

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

....architectures, such as ethernet and ATM, deliver relatively poor performance to real applications, despite impressive increases in raw bandwidth. This has been shown to be largely due to high software overhead. To address this problem a variety of user level network interfaces have been proposed[1, 2, 3, 4, 5, 6]. These have achieved dramatically reduced overhead and latency, and deliver substantial performance benefits to applications. However, these new interfaces are typically unsuitable for use in general purpose local area networks. The interface is often designed to support a particular class of ....

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.

....network interface and a network router. MDPs can be connected in a three dimensional grid to create an instance of the J Machine, a massively parallel architecture that scales up to thousands of nodes. The Mosaic is similar except that it is based on a 2 D rather than 3 D mesh. The Myrinet network [31] is constructed from arrays of Mosaic chips. The Henry Joerg and T network interfaces were integrated into the Motorola 88110, a commercial superscalar RISC processor. The advantage of a tightly coupled design is to be able to achieve the lowest possible latency by bringing data from the network ....

....to the interface card [36, 37, 38] This technique is called outboard buffering. Examples of network interfaces that attach to the I O bus are commercial Ethernet controllers, an ATM PCI interface called DART [39, 40] a PCI to PCI bridge called Memory Channel [41] the Myrinet network interface [31] and the Princeton SHRIMP network interface [42, 43] 2.1.2 Data movement The fundamental purpose of the physical interface is data movement. The send interface injects packets into the network and the receive interface ejects packets from the network. The two competing styles for data movement ....

[Article contains additional citation context not shown here]

Nanette J. Boden et al. Myrinet: a Gigabit-per-Second Local Area Network. IEEE Micro, February 1995. Also available on the World Wide Web through http://www.myri.com/research/index.html .

....attainable. Also, mixing static with dynamic checking is a promising research direction (cf. 15] iv) The actual implementation of the model into two available hardware architectures: 2 Quad Pentium Pro machines interconnected with Fast Ethernet, and 8 Dual Pentiums interconnected with Myrinet [2]. Acknowledgments. The first author would like thank Matthew Hennessy, Julian Rathke, and Kohei Honda for fruitful discussions. ....

Nanette J. Boden et al. Myrinet: 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, February 1995, pp. 29 - 36.

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-persecond local-area network. In IEEE Micro, volume 15, Arcadia, CA, February 1995. Myricom. http://www.myri.com.

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 gigabitper -second local-area network. IEEE Micro, 15(1):29--36, Feb 1995.

No context found.

Nanette J. Boden and Danny Cohen and Robert E. Felderman and Alan E. Kulawik and Charles L. Seitz and Jakov N. Seizovic and Wen-King Su, "Myrinet: 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--36, 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-per129 second local-area network. IEEE Micro, 15(1):29--36, February 1995. Available from http://www.myri.com/research/publications/Hot.ps.

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--36, 1995.

No context found.

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.

No context found.

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

No context found.

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.

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--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 Gigabitper -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 J. Seitz, Jakov N. Seizovic, and Wen-KingSu. Myrinet - A Gigabit-per-second Local-Area Network. IEEE Micro, 15(1), 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--36, February 1995.

First 50 documents  Next 50

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