Reschke, C., Sterling, T., Ridge,D., Savarse, D., Beer, D., Merkey P., "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation." Proceedings of the Fifth IEEE International Symposium on High Performance Distributed Computing 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....is the most critical part of a cluster. The networks of Xibalba are based on commodity 100 MBit s Fast Ethernet, like in most Beowulf class systems. Before inexpensive single backplane networking switches became readily available several different topologies were proposed for Beowulf clusters [16]. As a major difference to most other Beowulf clusters, Xibalba has two Fast Ethernet networks as specified below. For dedicated networks in parallel computing several high speed interconnect technologies were developed, e.g. with Myrinet. A comparison between two such technologies and a ....

Chance Reschke, Thomas Sterling, Daniel Ridge, Daniel Savarese, Donald Becker, and Phillip Merkey. A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation. In Proceedings of the Fifth IEEE International Symposium on High Performance Distributed Computing, pages 626--635. IEEE Comput. Soc. Press, Los Alamitos, CA, USA, 1996.

....critical part of a cluster. The networks of Xibalba are based on commodity 100 MBit s Fast Ethernet interconnects, like in most Beowulf class systems. Before inexpensive single backplane networking switches became readily available several different topologies were proposed for Beowulf clusters [16]. As a major difference to most other Beowulf clusters, Xibalba has two Fast Ethernet networks as specified below. For dedicated networks in parallel computing several high speed interconnect technologies were developed, e.g. with Myrinet. A nice comparison between two such technologies and a ....

Chance Reschke, Thomas Sterling, Daniel Ridge, Daniel Savarese, Donald Becker, and Phillip Merkey. A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation. In Proceedings of the Fifth IEEE International Symposium on High Performance Distributed Computing, pages 626--635. IEEE Comput. Soc. Press, Los Alamitos, CA, USA, 1996.

....of nodes that one can add. After this such limit, every additional node will not contribute processing power anymore and may even slow down the network. This paper attempts to discuss the hardware scalability issues of clusters and propose means in overcoming these issues. Based on some studies[4][8][5] it has become quite certain that a major problem when building large clusters is the underlying network interconnect. As clusters get larger and larger, maintainers simply add nodes in order to enhance cluster performance. Here are some of the issues that they might face with regards to ....

....configuration based on diameter and degree. For example, G(8; 1, 4) is already the largest graph of all graphs with diameter equals to 2 and regularity 3 [6] Use of wire speed switches. Wire speed switches are necessary for improved performance and for the maximization of network topology[8]. However, the problem is that it is not easy to determine if a switch is wire speed. It would be best to consult with the manufacturer for this information. Routing is non trivial. As seen in the routing procedure above, the routing of this network topology is non trivial. Each subnet will ....

Reschke, C., Sterling, T., Ridge,D., Savarse, D., Beer, D., Merkey P., "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation." Proceedings of the Fifth IEEE International Symposium on High Performance Distributed Computing 1996.

....with other workstations. One of the original Channel Bonding technique was the Bonded Dual Net where the network topology consists of a pair of Ethernet buses that spanned the entire computer cluster and operated in parallel as a single virtual bus having an aggregate bandwidth of 20Mbps. [9] To improve the Bonded Dual Net approach, Routed Mesh topology was used where computers were arranged in a mesh like fashion having a total of eight network segments. In the 16computer cluster, each computer is directly connected to three computers in the first Ethernet card while three more ....

....the transmitting computer must use one of the computers as a router. Another scheme used was the Switched Mesh where two Ethernet switches were added to provide routing between segments. The eight network segments were retained with four network segments connected to the each Ethernet switch. [9] Parallel Communication In Parallel Communication, multiple Ethernet networks were used to connect several SPARC stations. This system uses two parallel Ethernet networks. There were three algorithms experimented. The first channel allocation algorithm used was the round robin technique. The ....

Reschke, C. et. al., A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation, Proceedings High Performance and Distributed Computing, 1996. 160

....project was one of the first projects pushing this solution [21] PVM (Parallel Virtual Machine) 22] and MPI (Message Passing Interface) 23] libraries provide messaging and synchronization constructs that are needed for distributed parallel computing with NOW solutions. Projects like Beowulf [24] for Linux are continuing the distributed parallel 8 approach. Research operating systems, like Spring from Sun [25] focus on distributed computing via the network. All of these approaches have one thing in common, the use of a network as a communication media. While the network is very ....

Reschke C., Sterling T., Ridge D., Savarese D., Becker D., Merkey P., "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation,", Proceedings of the Fifth IEEE Symposium on High Performance Distributed Computing, 1996

....with sustained performance of Gigaflops. Rapid advances in PC microprocessor floating point performance, high bandwidth backplane and network standardized technologies, and open POSIX compliant systems software have created an important opportunity for a new business mode in high end computing[2]. Figure 1 even shows us an example of a Beowulf class machine built with old and unused PCs. While still experimental, Pile of PCs are emerging as an important new class of parallel computing system, complementing more expensive vendor driven medium level MPPs and SMPs, and providing new ....

....Ethernet networks (10 Mbps) to achieve useful sustained throughput gain. For example, dual Ethernet have delivered sustained bandwidth of almost 2 Mbytes per second under favorable conditions of packet size, This was achieved in a user transparent manner through changes in the operating system. [2] Beowulf employs the Linux operating system which, among its other features, comes with source code and therefore is ideal for this class of research. Modifications to Linux were made to support channel bonding, allowing message packets to use any of the available networks [5] A new realm of ....

C. Reschke, T. Sterling, D. Ridge, D. Savarse, D. Beer, P. Merkey. 1996. "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation." Proceedings of the Fifth IEEE International Symposium on High Performance Distributed Computing.

....hardware components, such as personal computers (PCs) and standard Ethernet adapters and switches. The original Beowulf cluster was developed at the Center of Excellence in Space Data and Information Sciences (CESDIS) under the sponsorship of the National Aeronautic Space Agency [4] 5] 6][7]. It employed 16 PC based processing modules integrated with multiple Ethernet networks. Large disk capacity and high disk to memory bandwidth was achieved through the use of hard disk and controller for each processing module supporting up to 16 way concurrent accesses. A series of experiments ....

C. Reschke, T. Sterling, D. Ridge, D. Savarse, D. Becker, P. Merkey. 1996. "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation." Proceedings of the Fifth IEEE International Symposium on High Performance Distributed Computing.

....project was one of the first projects pushing this solution [2] PVM (Parallel Virtual Machine) 3] and MPI (Message Passing Interface) 4] libraries provide messaging and synchronization constructs that are needed for distributed parallel computing with NOW solutions. Projects like Beowulf [5] for Linux are continuing the distributed computing approach. All of these approaches have one common resource, the network. While the network is very important, no strong push has been made to provide fault tolerance for network failures in a server cluster solution. We developed a network ....

C. Reschke, et. al, "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation,", IEEE High Performance Distributed Computing, 1996.

....environment like this offers the following advantages for testing purposes: low hard and software costs . simple administration . availability (the system can be dedicated to the test program so results are not influenced by changing workloads etc. Furthermore the Beowulf projects [1] [13] [16] have shown that due to some enhancements in the network topology such PC networks can provide peak performance in excess of 1 GOPS and also high disk bandwidth. Even this early ViPIOS implementation shows promising performance results. In order to test ViPIOS, we carried out several ....

Chance Reschke, Thomas Sterling, and Daniel Ridge, A design study of alternative network topologies for the beowulf parallel workstation, Proceedings, High Performance and Distributed Computing, 1996, 1996.

No context found.

Chance Reschke, Thomas Sterling, Daniel Ridge, Daniel Savarese, Donald J. Becker, and Phillip Merkey. A design study of alternative network topologies for the beowulf parallel workstation. In Proceedings of the Fifth High Performance Distributed Computing (HPDC '96), pages 626--636, August 1996.

....0 7695 0001 3 99 10.00 (c) 1999 IEEE Proceedings of the 32nd Hawaii International Conference on System Sciences 1999 9 for M 2 COTS clusters, and the design of a new topology, the P Mesh. Some of the early work on the Beowulf project also focused on evaluation of networking topologies [7]. The Beowulf project experimented with hubs as the interconnect for each row and column of a four by four mesh of PCs at the time, switching technology was not a commodity . The idea of using bridging hardware to interconnect rows and columns of a mesh of processors dates back as far as the ....

Reschke, Chance, Thomas Sterling, Daniel Ridge, Daniel Savarese, Donald Becker, Phillip Merkey, "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation", Proceedings, High Performance and Distributed Computing, 1996

No context found.

Chance Reschke, Thomas Sterling, Daniel Ridge, Daniel Savarese, Donald Becker, Philip Merkey, "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation," Proceedings, High Performance and Distributed Computing, 1996.

No context found.

C. Reschke, T. Sterling, D. Ridge, D. Savarese, and D. Becker. "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation,", Proceedings of the Fifth IEEE Symposium on High Performance Distributed Computing (HPDC), August 1996.

No context found.

Reschke, C., Sterling, T., Ridge,D., Savarse, D., Beer, D., Merkey P., "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation." Proceedings of the Fifth IEEE International Symposium on High Performance Distributed Computing 1996.

No context found.

Reschke, C., Sterling, T., Ridge,D., Savarse, D., Beer, D., Merkey P., "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation." Proceedings of the 5th IEEE International Symposium on High Performance Distributed Computing 1996.

No context found.

C. Reschke, T. Sterling, D. Ridge, D. Savarse, D. Becker, P. Merkey, "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation," Proceedings of the Fifth IEEE International Symposium on High Performance Distributed Computing, (1996).

No context found.

Reschke, C., Sterling, T., Ridge,D., Savarse, D., Beer, D., Merkey P., "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation." Proceedings of the 5th IEEE International Symposium on High Performance Distributed Computing 1996.

No context found.

C. Reschke, T. Sterling, D. Ridge, D. Savarese, D. Becker, P. Merkey. 1996. "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation." High Performance and Distributed Computing.

No context found.

Chance Reschke, Thomas Sterling, Daniel Ridge, Daniel Savarese, Donald Becker, and Phillip Merkey. A design study of alternative network topologies for the beowulf parallel workstation. In High Performance and Distributed Computing, 1996.

No context found.

Reschke, C.; Sterling, T.; Ridge, D.; Savarese, D.; Becker, D.; Merkey, P., "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation" in Proceedings High Performance and Distributed Computing, 1996.

No context found.

C. Reschke, T. Sterling, D. Ridge, D. Savarse, D. Becker, P. Merkey, "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation," Proceedings of the Fifth IEEE International Symposium on High Performance Distributed Computing, (1996).

No context found.

C. Reschke, T. Sterling, D. Ridge, D. Savarse, D. Beer, P. Merkey. 1996. "A Design Study of Alternative Network Topologies for the Beowulf Parallel Workstation." Proceedings of the Fifth IEEE International Symposium on High Performance Distributed Computing.

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