BCL-3: A high performance basic communication protocol for commodity superserver DAWNING-3000
This paper introduces the design and implementation of BCL-3, a high performance low-level communication software running on a cluster of SMPs (CLUMPS) called DAWNING-3000. BCL-3 provides flexible and sufficient functionality to fulfiil the communication requirements of fundamental system software developed for DAWNING-3000 while guaranteeing security, scalability, and reliability. Important features of BCL-3 are presented in the paper, including special support for SMP and heterogeneous network environment, semiuser-level communication, reliable and ordered data transfer and scalable flow control. The performance evaluation of BCL-3 over Myrinet is also given.
Keywordscluster communication system user-level communication scalable flow control CLUMPS
Unable to display preview. Download preview PDF.
- Hwang Kai, Xu Zhiwei. Scalable Parallei Computing: Technology, Architecture, Programming. WCB/McGraw-Hill, 1998.Google Scholar
- Hiroshi Tezuka, Francis O’Carroll, Atsushi Hori, Yutzka Ishikawa. Pin-down cache: A virtual memory management technique for zero-copy communication. InProc. 12th International Parallel Processing Symposium & 9th Symposium on Parallel and Distributed Processing. Florida, USA, 1998, pp. 308–314.Google Scholar
- Gropp W W, Lusk E L. A taxonomy of programming models for symmetric multiprocessors and SMP clusters. InProc. Programming Models for Massively Parallel Computers, Berlin, Germany, 1995, pp. 2–7.Google Scholar
- Luecke G R, Coyle J J. Comparing the performance of MPI on the Cray T3E-900, the Cray origin 2000 and the IBM P2SC.Electronic Journal of Performance Evaluation and Modelling for Computer Systems (PEMCS), 1998, available at URL: http://hpc-journals.ecs.soton.ac.uk/PEMCS/.Google Scholar
- T von Eicken, Culler D E, Goldstein S C, Schauser K E. Active messages: A mechanism for integrated communication and computation. InProc. the 19th Annual International Symposium on Computer Architecture, Gold Coast, Gueensland, Australia, 1992, pp. 256–266.Google Scholar
- Lumetta S S, Mainwaring A M, Culler D E. Multi-protocol active messages on a cluster of SMP’s InProc. Supercomputing 97, California, USA, 1997, electronic proceedings only, available at URL: http://www.supercomp.org/sc97/proceedings/.Google Scholar
- Lumetta S S, Culler D E. Managing concurrent access for shared memory active messages. InProc. the International Parallel Processing Symposium, Florida, USA, 1998, pp. 272–278.Google Scholar
- Prylli L, Tourancheau B. BIP: A new protocol designed for high-performance networking on Myrinet. InWorkshop PC-NOW, IPPS/SPDP98, Orlando, USA, 1998, pp. 472–485.Google Scholar
- Patrick Geoffray, Loïc Prylli, Bernard Tourancheau, BIP-SMP: High performance message passing over a cluster of commodity SMP’s. InProc. Supercomputing 99, 1999, electronic proceedings only, available at URL: http://www.supercomp.org/sc99/proceedings/.Google Scholar
- Loïc Prylli, Bernard Tourancheau, Roland Westrelin. An improved NIC program for high-performance MPI. InProc. ACM99, 1999, electronic proceedings only, available at URL: http://www.crhc.uiuc.edu/~steve/wcbc99/.Google Scholar