Measuring memory hierarchy performance of cache-coherent multiprocessors using micro benchmarks

ABSTRACT

Even with today's large caches, the increasing performance gap between processors and memory systems imposes a memory bottleneck for many important scientific and commercial applications. This bottleneck is intensified in shared-memory multiprocessors by contention and the effects of cache coherency. Under heavy memory contention, the memory latency may increase 2 or 3 times. Nonethless, as more sophisticated techniques are used to hide latency and increase bandwidth, measuring memory performance has become increasingly difficult. Previous simple methods to measure memory performance can overestimate uniprocessor memory latency and underestimate bandwidth by tens of percent. This paper introduces a micro benchmark suite that measures memory hierarchy performance in light of both uniprocessor optimizations and the contention and coherence effects of multiprocessors. The benchmark suite has been used to improve the memory system performance of the SGI Origin multiprocessor.

REFERENCES

• 1: Mike Galles. Scalable Pipelined Interconnect for Distributed Endpoint Routing: The SGI SPIDER chip. In Hot Interconnects '96.
• 2: John L. Hennessy , Norman P. Jouppi, Computer Technology and Architecture: An Evolving Interaction, Computer, v.24 n.9, p.18-29, September 1991 [doi>10.1109/2.84896]
• 3: John L. Hennessy , David A. Patterson, Computer architecture (2nd ed.): a quantitative approach, Morgan Kaufmann Publishers Inc., San Francisco, CA, 1996
• 4: Wilbur H. Highleyman, Performance analysis of transaction processing systems, Prentice-Hall, Inc., Upper Saddle River, NJ, 1989
• 5: David Kroft, Lockup-free instruction fetch/prefetch cache organization, Proceedings of the 8th annual symposium on Computer Architecture, p.81-87, May 12-14, 1981, Minneapolis, Minnesota, United States
• 6: James Laudon , Daniel Lenoski, The SGI Origin: a ccNUMA highly scalable server, Proceedings of the 24th annual international symposium on Computer architecture, p.241-251, June 01-04, 1997, Denver, Colorado, United States
• 7: Daniel E. Lenoski , Wolf-Dietrich Weber, Scalable Shared-Memory Multiprocessing, Morgan Kaufmann Publishers Inc., San Francisco, CA, 1994
• 8: John D. McCalpin. Sustainable Memory Bandwidth in Current High Performance Computers. http://reality.sgi.com/mccalpin_asd., October 1995.
• 9: Larry McVoy, Carl Staelin. lmbench: Portable tools for performance analysis. http://reality.sgi.com/lm_engr/index.html., 1995.
• 10: Mark S. Papamarcos , Janak H. Patel, A low-overhead coherence solution for multiprocessors with private cache memories, Proceedings of the 11th annual international symposium on Computer architecture, p.348-354, January 1984
• 11: Rafael Saavedra, R. Stockton Gaines, Michael Carlton. Characterizing the Performance Space of Shared Memory Computers Using Micro Benchmarks. July 1993.
• 12: P. Sweazey , A. J. Smith, A class of compatible cache consistency protocols and their support by the IEEE futurebus, Proceedings of the 13th annual international symposium on Computer architecture, p.414-423, June 02-05, 1986, Tokyo, Japan
• 13: The Ultra Enterprise 10000 Server. http://www.sun.com/servers/datacenter/products/starfire/wp.html.
• 14: Stephen J. Von Worley , Alan J. Smith, Microbenchmarking and Performance Prediction for Parallel, University of California at Berkeley, Berkeley, CA, 1995
• 15: Kenneth C. Yeager, The MIPS R10000 Superscalar Microprocessor, IEEE Micro, v.16 n.2, p.28-40, April 1996 [doi>10.1109/40.491460]