The Scalable Coherent Interface (SCI) as an Alternative for Cluster Interconnection
Resumo
The SCI standard was proposed in 1992 to be a high performance bus for processor interconnection in multicomputers. With the growing popularity of cluster architectures, the standard was implemented in PCI cards and of offered as a flexible an efficient alternative for the construction of parallel systems. After some initial difficulties, like incompatibilities with various PCI chipsets and the absence of drivers for some platforms, the SCl-based products are already mature but the expected breakthrough is not yet reached. In this paper we present our experience in the construction and utilisation of SCI clusters and in the development of support tools and applications for these systems. We present a survey of the obtained results over the last years in this area, including our own research, and make an analysis on how the standard and its implementations are performing today.
Referências
ÁVILA, R. B. et al. OptiSCI: a visual environment to optimize the placement of shared memory segments on a SCI cluster. In: SBAC, 12., 2000, São Pedro, SP. Proceedings ... São Carlos: UFSCAR, 2000. p. 129-135.
BARRETO, M.; ÁVILA, R.; NAVAUX, P. The MultiCluster model to the integrated use of multiple workstation clusters. In: PC-NOW, 3., 2000, Cancun. Proceedings ... Berlin: Springer, 2000. p.71-80. (Lecture Notes in Computer Science, v.1800).
BARRETO, M. E. DECK: um ambiente para programação paralela em agregados de multi processadores. Porto Alegre: PPGC da UFRGS, 2000. Dissertação de Mestrado.
BARRETO, M. et al. lmplementation of the DECK environment with BIP. In: MYRINET USER GROUP CONFERENCE, 1., 2000, Lyon, France. Proccedings . -- Lyon: LNRIA Rocquencourt, 2000. p.82-88.
BODEN, N. et al. Myrinet: a gigabit-per-second local-area network. lEEE Micro, Los Alamitos, v. 15, n.1, p.29-36, Feb. 1995.
DOLPHIN interconnect solutions web. Disponível por WWW em http://www.dolphinics.no (abr. 2000).
DORMANNS, M.; SCHOLTYSSIK, K.; BEMMERL, T. A shared-memory interface for SCI clusters. ln: HELLWAGNER, H.; REINEFELD, A. (Eds.). SCI: Scalable Coherent Interface: architecture and software for high-performance compute clusters. Berlin: Springer, 1999. p.490. (Lecture Notes in Computer Science. v. 1734).
EICKEN, T. von et al. Active messages: a mechanism for integrated communication and computation. In: ISCA. 19 .. 1992. Gold Coast, Australia. Proccedings .-. New York: ACM, 1992. p.256-266.
FISCHER, M.; REINEFELD. A. PVM for SCI clusters. In: HELLWAGNER, H.; REINEFELD, A. (Eds.). SCI: Scalable Coherent Interface: architecture and software for high-performance compute clusters. Berlin: Springer, 1999. p.239-248. (Lecture Notes in Computer Science, v.1734).
GIACOMINI, F. et al. Low-level SCI software requirements, analysis and predesign. (S.I.): ESPRIT Project 23174- Software lnfrastructure for SCI (SISCI), 1998.
GM. Disponível por WWW em http://www.myri.com/GM (dez. 1999).
HELLWAGNER, H.; REINEFELD, A. (Eds.). SCI: Scalable Coherent Interface: architecture and software for high-performance compute clusters. Berlin: Springer, 1999. 490p. (Lecture Notes in Computer Science, v.1734).
HERLAND, B. G.; EBERL. M.; HELLWAGNER, H. A common messaging layer for MPI and PVM over SCI. In: HPCN, 1998, Amsterdam. Proccedings ... Berlin: Springer-Verlag, 1998. p.576-587. (Lecture Notes in Computer Science, v.1401).
HUSE, L. P. et al. ScaMPI-design and implementation. In: HELLWAGNER, H.; REINEFELD, A. (Eds.). SCI: Scalable Coherent Interface: architecture and software for high-performance compute clusters. Berlin: Springer, 1999. p.249-261. (Lecture Notes in Computer Science, v.1734).
IBEL, M. et al. lmplementing Aclive Messages and Split-C for SCI clusters and some architectural implications. In: SCIZZL, 6., 1996, Santa Clara, USA. Proceedings ... [S.I.: s.n.], 1996.
lNSTITUTE OF ELECTRICAL ANO ELECTRONIC ENGINEERS. lEEE standard for scalable coherent interface (SCI). New York: [s.n.], 1992. IEEE 1596-1992.
MPI FORUM. the MPI message passing interface standard. Knoxville: University of Tennessee, 1994.
OLIVEIRA, F. A. D. de. Uma biblioteca para programação paralela por troca de mensagens de clusters baseados na tecnologia SCI. Porto Alegre: PPGC da UFRGS, 2001. Dissertação de Mestrado.
PRYLLI, L.; TOURANCHEAU, B. BIP: a new protocol designed for high-performance networking on Myrinet. In: PCNOW, 1., 1998. Proceedings ... Berlim: Springer, 1998. p.472-485. (Lecture Notes in Computer Science, v.I388).
SCHULZ, M. SCJ-VM: a tlexible base for transparent shared memory programming models on clusters of PCs. In: HIPS. 1999, San Juan, Puerto Rico. Proceedings ... Berlin: Springer, 1999. (Lecture Notes in Computer Science, v. 1586).
TASKJN, H.; BUTENUTH, R.; HEISS, H.-U. SCI for TCP/IP for Linux. In: SCI-EUROPE, 1998, Bordeaux, France. Proceedings ... [S.l.: s.n.], 1998.
TASKJN, H. Synchronisationsoperationen für gemeinsamen speicher in SCI-clustem. Paderbom: Universitat GH Paderbom. 1998. Diplomarbeit.
WORRINGEN, J. SCI-MPICH: the second generation. In: SCI EUROPE, 3., 2000. Munich, Germany. Proceedings ... [S.I.: s.n.], 2000. p. 10-20. Organizado como conference stream do Euro-Par'2000.
ZORAJA, I.; HELLWAGNER. H.; SUNDERAM, V. SCIPVM: parallel distributed computing on SCI workstation clusters. Concurrency: Practice and Experience, v. 11, n. 13, p. 121-138, Mar. 1999.