Manuals / AMD / Computer Equipment / Computer Hardware

AMD 64 Chapter, Analysis and Recommendations, Medium Total Time for both threads write-write

Models: 64

1 48

Download 48 pages 55.63 Kb

Page 33

Medium: Total Time for both threads (write-write)

40555 Rev. 3.00 June 2006

1.8

1.6

1.4

1.2

1

0.8

0.6

0.4

0.2

0

Performance Guidelines for AMD Athlon™ 64 and AMD Opteron™ ccNUMA Multiprocessor Systems

Medium: Total Time for both threads (write-write)

	146%	129%	129%	139%
				139%

0 Hop		0 Hop	0 Hop	0 Hop
0 Hop		1 Hop	1 Hop	2 Hop

0.0.w.0 0.1.w.0 (0 Hops) (0 Hops)

0.0.w.0 0.1.w.1 (0 Hops) (1 Hops)

0.0.w.0 0.1.w.2 (0 Hops) (1 Hops)

0.0.w.0 0.1.w.3 (0 Hops) (2 Hops)

Figure 13. Both Write-Only Threads Running on Node 0 (Different Cores) under Medium Background Load (High Subscription)

1.8

1.6

1.4

1.2

1

0.8

0.6

0.4

0.2

0

High: Total Time for both threads (write-write)

	146%	136%	146%
			135%
			135%

	0 Hop		0 Hop	0 Hop	0 Hop
	0 Hop		1 Hop	1 Hop	2 Hop

0.0.w.0 0.1.w.0 (0 Hops) (0 Hops)

0.0.w.0 0.1.w.1 (0 Hops) (1 Hops)

0.0.w.0 0.1.w.2 (0 Hops) (1 Hops)

0.0.w.0 0.1.w.3 (0 Hops) (2 Hops)

Figure 14. Both Write-Only Threads Running on Node 0 (Different Cores) under High Background Load (High Subscription)

As shown in Figure 15 below, the advantage disappears at very high loads and the 0 hop-1 hop case becomes slower than the 0 hop-0 hop case.

Chapter 3

Analysis and Recommendations

33

Image 33

AMD 64 manual Chapter, Analysis and Recommendations, Medium Total Time for both threads write-write

Contents

Performance Guidelines for AMD Athlon 64 and Issue Date JuneAMD Opteron ccNUMA Multiprocessor Systems Application NoteTrademarks 2006 Advanced Micro Devices, Inc. All rights reservedChapter 2 Experimental Setup ContentsContents Chapter 3 Analysis and RecommendationsA.2.1 What Resources Are Used When a Single Read-Only or List of Figures List of FiguresPerformance Guidelines for AMD Athlon 64 and AMD Opteron List of FiguresccNUMA Multiprocessor Systems 40555 Rev. 3.00 JuneRevision History Revision HistoryccNUMA Multiprocessor Systems Performance Guidelines for AMD Athlon 64 and AMD OpteronPerformance Guidelines for AMD Athlon 64 and AMD Opteron Revision HistoryccNUMA Multiprocessor Systems 40555 Rev. 3.00 JuneChapter 1 Introduction ChapterIntroduction 1.1 Related Documents 10 http//msdn2.microsoft.com/en-us/library/ms186255SQL.90.aspxIntroduction 14 http//msdn2.microsoft.com/en-us/library/tt15eb9t.aspx ChapterIntroduction Performance Guidelines for AMD Athlon 64 and AMD Opteron IntroductionccNUMA Multiprocessor Systems ChapterExperimental Setup Chapter 2 Experimental Setup2.1 System Used ChapterExperimental Setup Figure 1. Quartet TopologyFigure 2. Internal Resources Associated with a Quartet Node 2.2 Synthetic TestExperimental Setup ChapterExperimental Setup Data Access Rate Qualifiers2.3.1 X-Axis Display 2.3 Reading and Interpreting Test GraphsExperimental Setup Chapter2.3.2 Labels Used 2.3.3 Y-Axis DisplayExperimental Setup 3.1 Scheduling Threads Chapter 3 Analysis and Recommendations3.1.1 Multiple Threads-Independent Data Analysis and Recommendations3.1.2 Multiple Threads-Shared Data 3.2 Data Locality Considerations3.1.3 Scheduling on a Non-Idle System Analysis and RecommendationsAnalysis and Recommendations Chapter2 Hop 100%Analysis and Recommendations 3.2.1 Keeping Data Local by Virtue of first TouchAnalysis and Recommendations ChapterAnalysis and Recommendations afterwords no longer needs the data structure and if only one of the worker threads needs the data structure. In other words, the data structure is not truly shared between the worker threads3.3 Avoid Cache Line Sharing Threads access local data3.4 Common Hop Myths Debunked 3.4.1 Myth All Equal Hop Cases Take Equal TimeAnalysis and Recommendations Threads firing at each other crossfireNode 0 Core Node 1 Core Node 2 Core Node 3 Core ChapterAnalysis and Recommendations Analysis and Recommendations Next, we increase the number of background threads to six, running onChapter 3.4.2 Myth Greater Hop Distance Always Means Slower TimeAnalysis and Recommendations 216%Analysis and Recommendations Both threads access memory on node0 hop-0 hop case for the write-only threads ChapterAnalysis and Recommendations Analysis and Recommendations In addition, three background threads are running on nodes 1, 2 andAnalysis and Recommendations ChapterMedium Total Time for both threads write-write High Total Time for both threads write-writeAnalysis and Recommendations 3.5 Locks147% 158% 158%Analysis and Recommendations ChapterPerformance Guidelines for AMD Athlon 64 and AMD Opteron Analysis and RecommendationsccNUMA Multiprocessor Systems ChapterChapter 4 Conclusions ConclusionsChapter Conclusions Data placement tools can also come in handy when a thread needs more data than the amount of physical memory available on a node. Certain OSs also allow data migration with these tools or API. Using this feature, data can be migrated from the node where it was first touched to the node where it is subsequently accessed. There is a cost associated with this migration and it is not advised to use it frequently. For additional details on the tools and APIs offered by various OS for thread and memory placement refer to Section A.7 on pageA.1 Description of the Buffer Queues Appendix AFigure 16. Internal Resources Associated with a Quartet Node Appendix APage Appendix A A.2.3 What Role Do Buffers Play in the Throughput Observed?0 Hop-1 Hop Case on an Idle System for Write- Only Threads? A.4 Why Is 0 Hop-0 Hop Case Slower Than theA.5 Why Is 0 Hop-1 Hop Case Slower Than Controlling Process and Thread Affinity A.7.1 Support Under LinuxA.7.3 Support under Microsoft Windows A.7.2 Support under SolarisControlling Memory Affinity Appendix AA.8.1 Support under Linux A.8.2 Support under SolarisA.8.3 Support under Microsoft Windows Appendix A A.8.4 Node Interleaving Configuration in the BIOSccNUMA Multiprocessor Systems Performance Guidelines for AMD Athlon 64 and AMD OpteronAppendix A 40555 Rev. 3.00 June