H. Hellwagner and A. Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. Springer-Verlag, 1999.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....from PC clusters. In Jie Tao is a staff member of Jilin University in China and is currently pursuing her Ph.D at Technische Universitat Munchen in Germany. order to achieve the necessary NUMA global memory support, the IEEE standardized [1] system area network SCI (Scalable Coherent Interface) [5] is deployed. This interconnection technology, realized with the help of standard PCI adapter cards, enables a direct remote memory access to any physical memory location in the whole cluster with the help of remote memory mappings, thereby establishing a NUMA system. With a latency of less than 2 ....

H. Hellwagner and A. Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. SpringerVerlag, 1999.

....system software. 3.1 Framework Currently, the hardware monitor is designed for our NUMA characterized SMiLE PC clusters. SMiLE stands for Shared Memory in a Lan like Environment and it is a project [3] broadly investigates in SCI based cluster computing . SCI (Scalable Coherent Interface [1]) is an IEEE standardized [2] interconnection technology with extremely low latency and very high bandwidth. In order to explore shared memory programming on top of this architecture, a software framework, called HAMSTER [4] Hybrid dsm based Adaptive and Modular Shared memory archiTEctuRe) is ....

H. Hellwagner and A. Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. Springer Verlag, 1999.

....system software. 3.1 Framework Currently, the hardware monitor is designed for our NUMA characterized SMiLE PC clusters. SMiLE stands for Shared Memory in a Lan like Environment and it is a project [3] broadly investigates in SCI based cluster computing . SCI (Scalable Coherent Interface [1]) is an IEEE standardized [2] interconnection technology with extremely low latency and very high bandwidth. In order to explore shared memory programming on top of this architecture, a software framework, called HAMSTER [4] Hybrid dsm based Adaptive and Modular Shared memory archiTEctuRe) is ....

H. Hellwagner and A. Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. Springer Verlag, 1999.

....the monitor simulator can stop and restart the monitoring during the simulated execution of programs. This enables the acquisition of per SCI (Scalable Coherent Interface) an IEEE standardized [8] interconnection technology with high bandwidth, low latency, and global physical access space [4]. phase information which can be used to optimize applications with dynamic changing access patterns. 6 Validation SIMT is a tool for evaluating shared memory systems. In order to examine its accuracy, we have compared the simulated execution of several applications with the execution on the ....

H. Hellwagner and A. Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science, chapter 3. Springer-Verlag, 1999.

....adaptation heuristics. 3. Target system ARS is currently established on top of the NUMA characterized SMiLE like PC clusters. SMiLE stands for Shared Memory in a LAN like Environment [15] and is a project broadly investigating in SCI based cluster computing. SCI (Scalable Coherent Interface) [4] is an IEEE standardized [7] interconnection technology with extremely low latency ( 2 #s) and very high bandwidth ( 300MB s for modern PC architectures) In addition, SCI provides a single physical address space across processor nodes in hardware, supporting the establishment of a distributed ....

H. Hellwagner, A. Reinefeld (Eds.), SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, Lecture Notes in Computer Science, vol. 1734, Springer, Berlin, 1999.

....the dynamic behavior of shared memory applications, and optimize the program with an application speci c data layout resulting in signi cant performance improvements. The approach has been developed and evaluated on PC clusters with an SCI interconnection technology (Scalable Coherent Interface [3, 5]) SCI supports memory oriented transactions over a ringlet based network topology, e ectively supporting a distributed shared memory in hardware. In order to support shared memory programming on top of such a NUMA architecture, a software framework has been developed within the SMiLE project ....

....environments, a software framework is required which closes the semantic gap between the distributed physical memories of NUMA machines and the global virtual memory abstraction required by shared memory programming models. Such a shared memory framework based on SCI (Scalable Coherent Interface [3]) an IEEE standardized [5] cluster interconnect with NUMA capabilities, has been developed within the SMiLE [7] Shared Memory in a LAN like Environment) project, which broadly investigates in SCI based cluster computing. In addition, this framework, called HAMSTER [13] Hybrid dsm based ....

H. Hellwagner and A. Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. Springer Verlag, 1999.

....approach we use a meta cluster at our university (ETH Zurich) that unites several PC clusters, connecting their interconnects to a dedicated cluster backbone. This cluster testbed offers a variety of topologies and networking speeds. The networks include some Gigabit networking technology like SCI [2,3] and Myrinet [4] with an emphasis on Fast and Gigabit Ethernet [5] The evaluation work was performed on the Patagonia sub cluster of 24 Dell 410 Desktop PCs configured as workstations with keyboards and monitors. The Intel based PC nodes are built around a dual Pentium II processor configuration ....

Hellwagner H, Reinefeld A (eds.) SCI: Scalable Coherent Interface---Architecture and Software for High-Performance Computer Clusters (Lecture Notes in Computer Science, vol. 1734). Springer: Berlin, 1999.

....about the lowlevel memory transactions, we have designed and implemented a hybrid performance monitor [5] Unlike the two hardware monitors mentioned above, our monitoring facility is used as the base of program tuning. The hardware monitor is implemented for our SCI based PC clusters. SCI 1 [3] is an IEEE standardized [1] System Area Network, which has a latency of less than 2 s and a bandwidth of more than 80 MB s for process to process communication. In addition, SCI offers a 64 bit physical address space accessible from any node. Shared memory programming can be enabled on the SCI ....

Hermann Hellwagner and Alexander Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. Springer-Verlag, 1999.

....round robin (size=1) optimized Figure 3. Impact of the data locality optimization on simulation time. In the next line of this research work, this kind of optimization will be validated on actual NUMA characterized systems, like PC clusters interconnected via the Scalable Coherent Interface [2], an IEEE standardized System Area Network with a latency of less than 2us and a bandwidth of more than 80 MB s for process to process communication. This work has begun with a Data Layout Visualizer [14] which illustrates the memory access histogram provided by the simulation tool in a ....

Hermann Hellwagner and Alexander Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. Springer-Verlag, 1999.

....a LAN like Environment) 1] at LRR TUM investigates high performance cluster computing. It is based on a NUMA architecture built from PC based clusters. In order to achieve a high communication performance, the high speed IEEE standardized [2] system area network SCI (Scalable Coherent Interface) [3] is deployed as the interconnection technology. With a latency of less than 2 and a bandwidth of far beyond 200 MB s for process to process communication, SCI offers state of the art performance for clusters and is therefore regarded as well suited for High Performance Computing. Within the ....

Hermann Hellwagner and Alexander Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. Springer-Verlag, 1999.

....important and are regarded as adequate architectures for High Performance Computing. Such clusters based on PCs have been built for our SMiLE project (Shared Memory in a Lan like Environment) 8] In order to achieve the necessary NUMA global memory support, the Scalable Coherent Interface (SCI) [1, 5], an IEEE standardized System Area Network, is taken as the interconnection technology. With a latency of less than 2 s and a bandwidth of more than 80 MB s for process to process communication, SCI offers stateof the art performance for clusters. Shared memory programming is enabled on the ....

Hermann Hellwagner and Alexander Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. Springer-Verlag, 1999.

....standard LANs restricts performance of parallel applications. To Jie Tao is a staff member of Jilin University of China and is currently pursuing her ph.D at Munich University of Technology of Germany overcome this performance gap the high speed low latency Scalable Coherent Interface (SCI) [4] is taken as interconnection technology. With hardware Distributed Shared Memory (DSM) and high performance communication characteristics a PC cluster is built for High Performance Computing(HPC) Secondly, fast data access and communication amongst an application s cooperating threads needs the ....

Hermann Hellwagner and Alexander Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. Springer-Verlag, 1999.

....with the use of a pool of threads, as used in Nexus and other environments, which will hopefully reduce the creation overhead. Other research topics within the project include the implementation on top of a different Myrinet protocol, such as GM, and on other communication networks, such as SCI [8]. Acknowledgements This work has been partially supported by grants from CNPq, CAPES and UFRGS. The authors would like to thank Patrick Geoffray, from the BIP team, for his help in installing and using BIP. ....

H. Hellwagner and A. Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for HighPerformance Compute Clusters, volume 1734 of Lecture Notes in Computer Science. Springer, Berlin, 1999.

....might have to wait for data requested from another computer. Therefore special care should be taken in choosing the network connection depending on the application requirements. Special network connections have been developed especially for clusters in order to obtain high bandwidth and low latency[11]. The ideal size of data to process in parallel is highly dependent on the type of network infrastructure and communication protocol. In each case, the transfer rate and the latency should be estimated by experiments for different size of data. Such graphs are already available for many ....

Hermann Hellwagner and Alexander Reinefeld (eds), SCI - Scalable Coherent Interface: Architecture and Software for High-Performance Compute Clusters, vol. 1734, Springer State-of-the-Art Surveys, Lecture Notes in Computer Science, 1999.

....1 Shared Memory in a LAN like Environment The SMiLE project [9] investigates high performance cluster computing. As the limited communication performance over standard LANs restricts performance of parallel applications, the high speed low latency Scalable Coherent Interface (SCI) 1] [8] is taken as the interconnection technology. SCI supports memory oriented transactions over a ringletbased network and enables a hardware supported distributed shared memory (DSM) system on top of a cluster of PCs. With hardware based DSM and high performance communication characteristics, the ....

....memory which is used to produce the packets that would be transfered if the application ran truly on our SCI Cluster. The packets are stored in a trace file, which is used to stimulate the hardware monitor simulation. B. Simulation of SCI VM and Cache Memory SCI Virtual Memory (SCI VM) 16] [8] is a concept for the implementation of a global virtual memory using hybrid hardware software DSM. This concept can be applied in a fully transparent manner allowing the execution of shared memory applications on a global virtual memory without any changes. The memory is distributed at page ....

[Article contains additional citation context not shown here]

Hermann Hellwagner and Alexander Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. Springer-Verlag, 1999.

....Memory in a LAN like Environment) project at LRR TUM. The SMiLE project [2,9] investigates in high performance cluster computing. As the limited communication performance over standard LANs restricts performance of parallel applications, the high speed low latency Scalable Coherent Interface(SCI) [1,5] Jie Tao is a staff member of Jilin University of China and is currently pursuing her Ph.D at Technische Universitat Munchen of Germany is taken as interconnection technology. With hardware Distributed Shared Memory (DSM) and high performance communication characteristics SCI based PC clusters ....

Hermann Hellwagner and Alexander Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. Springer-Verlag, 1999.

....means to support the administrator and focus on the modifications made to efficiently operate such clusters. The last chapter summarizes the work presented here. 2 SCI Clusters at PC # In 1996 we started with the first tests to build an SCI (scalable coherent interface, IEEE Std. 1596 1992 [15]) based PC cluster. We evaluated about half a dozen chip sets for their utilization of the PCI bus. Based on the results of this first, heterogeneous cluster the PSC 64 has been installed in December 1997 at PC # . The system manufactured by Siemens Nixdorf and SCALI [26] comprises of 32 ....

....and the status of the nodes (e.g allocated, free, defect, etc. Allocated partitions can be highlighted and node specific information can be achieved by clicking on a node symbol. Figure 7 depicts the ccsView tool. 6 Figure 7. CCS partition monitoring with ccsView. The second one, ccsMon [15], gives a more physical view on the cluster and the status of the SCI nodes. On each node, a local monitoring agent is sampling data and sends it periodically to a central server daemon. The server makes this information available to the graphical user inter interfaces, shown in Figure 8. The CCS ....

H. Hellwagner, A. Reinefeld (eds.). SCI:Scalable Coherent Interface: Architecture and Software for High-Performance Compute Clusters Lecture Notes in Computer Science 1734, 1999.

....transputer system with a 32x32 3 grid of T805 links, a 64 processor PowerPC 604 system with a fat grid topology, and a 48 node PowerPC system with a mesh of Clos topology of HIC (IEEE Std. 1355 1995) interconnects. CCS is also used for the management of two PC clusters. Both have a fast SCI [21] network with a 2D torus topology. The larger of the two clusters is a Siemens hpcLine [22] with 192 Pentium II processors. It has all typical features of a dedicated high performance computer and is therefore operated in multi user space sharing mode. The hpcLine is also embedded in the EGrid ....

....tries to use as few rings as possible (thereby minimizing the number of dimension changes from X to Y ringlets) while the second tries to map applications on single rings to ensure maximum bandwidth. Note that the best choice of policy also depends on the network protocol, e.g. in our case SCI [21]. The API of the MM allows to implement mapping modules that are optimally tailored to the specific hardware properties and network topology. As a side effect of this model, it is also possible to migrate partitions when they are not active. This feature is used to improve the utilization on ....

[Article contains additional citation context not shown here]

H. Hellwagner, and A. Reinefeld (eds.). SCI:Scalable Coherent Interface: Architecture and Software for HighPerformance Compute Clusters. Springer Lecture Notes in Computer Science 1734, 1999.

No context found.

H. Hellwagner and A. Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. Springer-Verlag, 1999.

No context found.

H. Hellwagner and A. Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. SpringerVerlag, 1999.

No context found.

H. Hellwagner and A. Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Computer Clusters, volume 1734 of Lecture Notes in Computer Science. Springer-Verlag, 1999.

No context found.

H. Hellwagner and A. Reinefeld, SCI: Scalable Coherent Interface Architecture and Software for High-Performance Compute Clusters. Berlin, Germany: Springer Verlag, 1999, pp. 89--101.

No context found.

Hermann Hellwagner and Alexander Reinefeld, editors. SCI: Scalable Coherent Interface: Architecture and Software for High-Performance Compute Clusters, volume 1734 of Lecture Notes in Computer Science. Springer, Berlin, 1999.

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