HPCN-Europe 1998: High-Performance Computing and Networking pp 638-647 | Cite as
Effects of network bandwidth on performance in software DSM systems
Abstract
Several studies have been done on the use of small-scale shared memory multiprocessor workstations as the primary building components for software distributed shared memory systems. None of the studies show performance which scales with the number of processors per workstation. The lack of scalability has been attributed to decreased per-processor network bandwidth with an increase in the number of processors per workstation.
We study the effects of network bandwidth on the performance of a software DSM system, Hemingway, in the context of SMP workstations. We find that network bandwidths do not need to scale by the number of processors — significant gains take place by just doubling the bandwidth even when increasing the number of processors per network by a factor of eight. Our results further indicate that poor system performance is due to the limitations by network latencies. Reducing network latencies, coupled with a small increase in network bandwidth produce the best results.
Keywords
Shared Memory Network Bandwidth Average Latency Network Latency Remote MemoryPreview
Unable to display preview. Download preview PDF.
References
- 1.A. Aggarwal and D. Grunwald. A performance evaluation of the Hemingway DSM system on a network of SMPs. CU-CS 837-97, Univ. of Colorado, Boulder, 1997.Google Scholar
- 2.A. Aggarwal, D. Grunwald, T. Hein, and E. Nemeth. Hemingway, a distributed shared memory system. CU-CS 813-96, Univ. of Colorado, Boulder, 1996.Google Scholar
- 3.M. A. Blumrich, K. Li, R. Alpert, C. Dubnicki, E. W. Felton, and J. Sandberg. Virtual memory mapped network interface for the shrimp multiprocessor. In Proceedings of the 21st Annual International Symposium on Computer Architecture, pages 142–154, April 1994.Google Scholar
- 4.Digital Equipment Corperation, Maynard, Mass. DECchip 21064 Microprocessor: Hardware Reference Manual, October 1992.Google Scholar
- 5.A. Erlichson, B. A. Nayfeh, J. P. Singh, and K. Olukotun. The benefits of clustering in shared address space multiprocessors: An applications-driven investigation. Technical Report CSL-TR-94-632, Stanford Univ., 1994.Google Scholar
- 6.R.B. Gillett. Memory Channel Network for pci. IEEE Micro, 16(1):12–18, February 1996.Google Scholar
- 7.L. Iftode, J. P. Singh, and K. Li. Understanding application performance on shared virtual memory systems. In Proceedings of the 23rd Annual International Symposium on Computer Architecture, pages 122–133, 1996.Google Scholar
- 8.A. Paithankar. AINT: A tool for simulation of shared-memory multiprocessors. Master's thesis, Department of Computer Science, Univ. of Colorado at Boulder, December 1995.Google Scholar
- 9.Daniel Scales, Kourosh Gharachorloo, and Anshu Aggarwal. Fine-grain software distributed shared memory on smp clusters. In Proceedings of the 4th Conference on High Performance Computer Architecture, February 1998.Google Scholar
- 10.R. Stets,S. Dwarkadas,N. Hardavellas,G. Hunt,L. Kontothanassis, S. Parthasarthy, and M. L. Scott. Cashmere-21: Software coherent shared memory on a clustered remote-write network. In Proceedings of the 16th ACM Symposium on Operating Systems Principles, October 1997.Google Scholar
- 11.J. E. Veenstra and R. J. Fowler. MINT: A front end for efficient simulation of shared-memory multiprocessors. In Proceedings of the Second International Workshop on Modeling, Analysis, and Simulation of Computer and Telecommunication Systems (MASCOTS), pages 201–207, 1994.Google Scholar
- 12.S. C. Woo, M. Ohara, E. Torrie, J. P. Singh, and A. Gupta. The SPLASH-2 programs: Characterization and methodological considerations. In Proceedings of the 22nd Annual International Symposium on Computer Architecture, pages 24–36, 1995.Google Scholar