AUTHORS: Allam Abumwais, Abdulkarim Ayyad
Download as PDF
ABSTRACT: A symmetric multi-core processor is a chip which integrates a number of processors (cores). Each core has its own local memory which is accessible by its core only. The cores share and equally access a shared memory. The multi-core processors suffer from the delay caused by the contention among the cores to access the shared memory. Also, a bigger delay results from the cache coherence operations, where each core must update other cores on any change it makes on a shared variable. This is accomplished by broadcasting the change to the rest of the cores. In 2014 the authors completed, tested and verified an organization of a multi-port content addressable memory (MPCAM) which, if used as a shared memory, it allows all the cores of the processor to access it simultaneously without the need for queuing and arbitration. The access time is the same as that of accessing the core’s private memory. The organization of this memory guarantees the cache coherence automatically and eliminates the need for cache coherence operations. This is an unprecedented result. This architecture represents a whole solution to the long standing problem of latency due to contention and cache coherence operations in multi-core (formerly multiprocessor) system
KEYWORDS: Multi-core, shared cache, contention, cache coherence, dual port CAM, multi-port content addressable memory (MPCAM).
REFERENCES:
[1] John L. Hennessy and David A. Patterson, “Computer Architecture A Quantitative Approach”, Fourth Edition Standford University, 2007.
[2] A. Ayyad, I. Exman, M. Land, L. Rudolph, “An Experimental Cross-Bar Switch For Support Of Collective Communications In Parallel Processing”, Proceedings of the Nineteenth Convention of IEEE in Israel, November 5-6, 1996.
[3] Barry Wilkinson, “Computer Architecture; Design and Performance”, 2nd edition, Prentice Hall Europe, 1996.
[4] J. Archibald and J.L. Baer, “Cache Coherence Protocols: Evaluation Using a Multiprocessor Simulation Model”, ACM transaction on Computer System, Vol 4, No.4, PP 273-298, 1986.
[5] B. Baterman, C. Freeman nad E. Resse, “A 450MHz 512KB second-Level cache with a 3,9GB/s data bandwidth ”, 1998.
[6] Daniel Hackenberg Daniel Molka Wolfgang E. “Nagel Comparing Cache Architectures and Coherency Protocols on x86-64 Multicore SMP Systems”, Center for Information Services and High Performance Computing (ZIH), MICRO’09, December 12–16, 2009, New York, NY, USA.
[7] “An Introduction to the Intel QuickPath Interconnect”, Intel Corporation, January 30, 2009.
[8] Abdulkarim Ayyad, “The Design of a Special Purpose Dual Port Content Addressable Memory”, Computer Architecture Lab Course Project Report, Al-Quds University, Palestine, 2011.
[9] Raymong Leong, Gary Green, “Dual-port content addressable memory”, Assignees: Cypress Semiconductor Corporation, Patent number: US6122706, Application number: 08/172,575, Filing date: Dec 22, 1993, Issue date: Sep 19, 2000.
[10] The homepage of Altera. http://www.altera.com/
[11] Altera, “Stratix III Development Kit”, Document Version: 1.1, San Jose, CA 95134,Agest 2008 (www.altera.com)
[12] Altera ,”Stratix III 3SL150 Development Board ”, Refernce Manual , San Jose, CA 95134,May 2013.
[13] Michael E. Thomadakis, Ph.D.,“The Architecture of the Nehalem Processor And Nehalem-EP SMP Platforms”, Supercomputing Facility, Research Report Texas A&M University, March, 17, 2011. miket@tamu.edu 2014.
[14] Robert Franz and Josef Weidendorfer, “Development of a Multicore Cache Simulator for Performance Analysis”, bachelor thesis, Technical University Munich , july 2008.
[15] The homepage of Valgrind. http://www.valgrind.org