OptiSCI: a Visual Environment to Optimize the Placement of Shared Memory Segments on a SCI Cluster
Resumo
The Scalable Coherent Interface (SCI) is an IEEE interconnection standard which is becoming widely used for the construction of parallel clusters of workstations. SCI provides a hardware-supported common address space shared by the computing nodes, enablin the use of shared-memory as a distributed communication mechanism. Due to the strong NUMA characteristic of SCI-based clusters, the placement of shared segments among the nodes has to be carefully planned, in order to minimize the cost of network communication latencies. In this paper we present a visual tool, OptiSCI, to help in the task of placing shared memory segments onto a SCI cluster with respect to such communication cost. OptiSCI is composed of a graphical modelling tool and a hardware simulator which makes use of a detailed cost model of SCI clusters in order to produce reliable results. At the end the tool is validated against a real implementation of a parallel application on a Linux SCI cluster.
Referências
Roger Butenuth and Hans-Ulrich Heiss. Shared memory programming on PC-based SCI clusters. In Hermann Hellwagner and Alexander Reinefeld, editors, Proc. of SCI-Europe ’98, Bordeaux, France, September 1998.
Rajkumar Buyya, editor. High Performance Cluster Computing: Architectures and Systems. Prentice Hall PTR, Upper Saddle River, 1999.
The Dolphin SCI interconnect. Available at [link], 1996.
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.
Roberto A. Hexsel. A Quantitative Performance Evaluation of SCI Memory Hierarchies. PhD thesis, University of Edinburgh, October 1994.
Kai Hwang and Zhiwei Xu. Scalable Parallel Computing: Technology, Architecture, Programming. McGraw-Hill, Boston, 1998.
IEEE. IEEE standard for scalable coherent interface (SCI). IEEE 1596-1992, 1992.
Uwe Kastens et al. Streets. Abschlußbericht, Universität Paderborn, 1998.
João Frederico Lacava Schramm. Ambiente gráfico para o desenvolvimento de aplicações distribuídas. Master’s thesis, CPGCC/UFRGS, 1996.
Thomas L. Sterling, John Salmon, Donald J. Becker, and Daniel F. Savarese. How to Build a Beowulf: a Guide to the Implementation and Application of PC Clusters. MIT, Cambridge, 1999.
Hüseyin Taşkın. Synchronisationsoperationen für gemeinsamen speicher in SCI-clustern. Diplomarbeit, Universität GH Paderborn, Paderborn, 1998.
