Return to Basics II : Understanding POWER7 Capacity Entitlement and Virtual Processors VN212 Rosa Davidson

Transcription

1 2011 IBM Power Systems Technical University October Fontainebleau Miami Beach Miami, FL Return to Basics II : Understanding POWER7 Capacity Entitlement and Virtual Processors VN212 Rosa Davidson Advanced Technical Skills - Security and Performance - IBM Copyright IBM Corporation 2012 Materials may not be reproduced in whole or in part without the prior written permission of IBM. 5.3

2 Session Evaluations ibmtechu.com/vp Prizes will be drawn from Evals Copyright IBM Corporation

3 Bibliography - References Beyond this presentation, read the White Paper from Mala Anand : POWER7 Virtualization - Best Practice Guide Server virtualization with IBM PowerVM IBM Systems Workload Estimator or IBM System p Advanced POWER Virtualization Best Practices Redbook: Virtualization Best Practice: Configuring Processor Resources for System p5 Shared-Processor Pool Micro-Partitions: An LPAR Review: Virtualization Tricks: A Comparison of PowerVM and x86-based Virtualization Performance: IBM Integrated Virtualization Manager: Achieving Technical and Business Benefits through Processor Virtualization: Java Performance Advisor is available ftp://ftp.software.ibm.com/aix/tools/perftools/jpa/aix61/ VIOS Performance Advisor is available Virtualization Performance Advisor is in develeopment (expected Q1/2012) 3

4 Summary of Return to Basics - Part I : Numbers of VPs Shared LPAR 20 Virtual Processors Shared Pool 10 Cores For each shared LPAR : The number of VPs MUST BE LESS THAN The number of CORES of the SHARED POOL. Frame in Shared LPARs 40 VPs LPAR 10 VPs LPAR 10 VPs There is a ratio between VPs/Cores : Higher the ratio, Less Uncapped capacity is available. LPAR 10 VPs Shared Pool 10 Cores LPAR 10 VPs While migrating to POWER7 servers, Cores have been reduced. VPs have been maintained. Thus, ratio values increase. Thus, Uncapped capacity is reduced. 4

5 Summary of Return to Basics - Part I : Capacity Entitlement CE = VP C O R E Capacity Entitlement means : 1 st, Entitlement Entitlement of time access to the core Running on Uncapped is Running on your ratio VPs/Cores. You plan to be dispatched on entitled capacity or on uncapped? vp20 vp20 vp20 vp12 vp12 vp12 vp20 vp20 vp20 vp20 vp20 vp12 vp12 vp12 vp12 vp12 This ratio evolves : vp11 vp11 vp11 vp11 vp11 vp11 vp11 vp11 vp10 vp10 vp10 vp10 vp10 vp10 vp10 vp With frame configuration changes. Thus : CE/VP RATIO Uncapped dependency A 0.95 B Performance on Uncapped is Performance less predictable. % Uncapped capacity must remain reasonable to limit this bad perf. dependency 5

6 Outline Part 2 Summary of Part 1 PowerVM: The Right Choice for the LPAR Type (20 slides) Dedicated : to be or not to be, Considerations Dedicated Donating: Considerations Shared LPAR: Uncapped Considerations VIO Server: Considerations Shared LPAR: LPAR (CE, VP) - the Choices Shared LPAR not covering Production Need. Shared LPAR for current Production Need. Shared LPAR : Wrong Idea #1 / Shared LPAR : Good Idea #1 Shared LPAR : Wrong Idea #2 / Shared LPAR : Good Idea #2 Performance and Advisors Summary of our recommendations Conclusions FAQs 6

7 Making the right choice for the LPAR Type To be or not to be * Dedicated LPAR 1 2 VP Dedicated POWERVM Hypervisor Dispatch Wheel = 10 ms used idle VP 30 VP 31 0 Core 0 Core 1 Core 2 Core 3 Core 4 Core 5 Core 6 Core 7 vp10 vp11 vp12 vp20 Core 8 Core 9 Core A vp52 Core B vp53 LPAR 2 2 VP Dedicated 1 vp20 vp52 vp used VP 40 VP 41 2 vp20 vp52 vp idle 3 vp52 vp53 Should IDLE be hunted? 4 5 vp52 vp52 vp53 vp53 LPAR 1 : NO % ratio (2 cores) 6 vp52 vp53 You keep 33% (25/75) as margin 33% is «your good» idle margin: Unplanned Peak Activity. Short Term Activity Growth Margin vp52 vp52 vp53 LPAR 2 : YES - 35/63% ratio (2 cores) Let s keep a margin: *33% = core Shared Pool Dedicated Cores Dedicated Donating Cores Either reduce VP if possible (parallelism need). Either change LPAR Type: release the idle. 7 * Known as nearly a copyright of William Shakespeare

8 Making the right choice for the LPAR Type Dedicated Considerations used idle LPAR 1 2 VP Dedicated VP 30 VP 31 0 POWER Hypervisor Dispatch Wheel = 10 ms Core 0 Core 1 Core 2 Core 3 Core 4 Core 5 Core 6 Core 7 Core 8 Core 9 Core A Core B vp10 vp11 vp12 vp20 vp52 vp53 LPAR 2 2 VP Dedicated 1 vp20 vp52 vp used VP 40 VP 41 2 vp20 vp52 vp idle 3 vp52 vp53 (+) PROCESSOR AFFINITY CPU, CACHE, : ALL is YOURS! Each VP knows its core number (1:1). No Core Reload between dispatches vp52 vp52 vp52 vp53 vp53 vp53 vp53 (+) LOW HYPERVISOR ACTIVITY 8 vp52 THE LOWEST RATE! 9 vp52 No brain to pick a core to dispatch a VP. No Core Reload to do. Shared Pool Dedicated Cores Dedicated Donating Cores 8 (-) BUT IDLE CYCLE WASTED: so be AWARE CHOOSE THE RIGHT ONES (like LPAR1). consistent and constant Usage LPARs are good candidates.

9 Making the right choice for the LPAR Type Dedicated Donating Considerations used idle LPAR 1 2 VP Dedicated VP 30 VP 31 POWER Hypervisor Dispatch Wheel = 10 ms Core 0 Core 1 Core 2 Core 3 Core 4 Core 5 Core 6 Core 7 Core 8 Core 9 Core A Core B 0 vp10 vp11 vp12 vp20 vp60 vp used LPAR 2 2 VP Dedicated Donating VP 40 VP vp20 vp20 vp60 vp60 vp61 vp61 3 vp60 vp61 (+ / -) PROCESSOR AFFINITY 4 vp60 vp61 CPU, CACHE, : NOT ALWAYS YOURS! Each VP knows its core number (1:1). Processor Reload between LPAR dispatches. (+ / -) HYPERVISOR ACTIVITY A LOW RATE only for SMALL DONORS! No brain to pick a core to dispatch a VP. phyp reloads Processor content between LPARs. (+) LPAR has the Idle Entitlement w/o Consumption Lengthy loaded/lengthy Idle LPARs are good candidates Shared Pool vp60 vp60 vp60 vp60 vp60 Dedicated Cores vp61 vp61 vp61 vp61 vp61 Dedicated Donating Cores Avoid Bipolar LPARs, i.e, rapidly changing profile LPARs. 9 Put them in dedicated or shared but at Capped on Peak.

10 Making the right choice for the LPAR Type Shared LPAR Uncapped Considerations planned - CE extra - >CE LPAR 1 3 VP Shared CE 0.3 VP 10 VP 11 LPAR 2 1 VP Shared CE 0.3 VP 20 physc= Share Pool Core 0 Core 1 Core 2 Core 3 Core 4 Core 5 vp10 vp11 vp12 vp20 VP 12 LPAR 3 2 VP Shared CE 1 VP 30 VP 31 1 vp20 2 vp20 physc=1.2 physc=1.3 3 VP SHAR = 10 CE SHAR = 3.6 LPAR 4 2 VP Shared CE 1 LPAR 5 2 VP Shared CE 1 VP 40 VP 41 VP 50 VP 51 physc=1.2 physc= vp20 vp10 vp11 vp12 7 vp20 8 vp10 vp11 vp12 vp20 9 vp10 vp11 vp12 vp20 10 UNCAPPED CAPACITY DOES ADD (-) HYPERVISOR ACTIVITY which core is free at demand time? MAY LOSE (-) PROCESSOR AFFINITY If you want the same core, you d plan for it!.. what we call CE MOST AFFECTED LPARS: the ones with a CE/VP at 0.1 if they rely on uncapped to GROW BIG Shared Pool CO SHAR = 6 APP min= 0

11 Making the right choice for the LPAR Type VIO Server Considerations VP SHAR = 10 CE SHAR = 3.6 VIO SERVER 3 VP CE0.3 VP 10 VP 11 VP 12 Data Bridges physc=1.2 LPAR 2 1 VP Shared CE 0.3 VP 20 LPAR 3 2 VP Shared CE 1 VP 30 VP 31 LPAR 4 2 VP Shared CE 1 LPAR 5 2 VP Shared CE 1 VP 40 VP 41 VP 50 VP 51 physc=1.2 physc= 0.6 physc=1.3 physc=1.2 VIO SERVERS are Data BRIDGES for the FRAME. HIGHLY AFFECTED BY DISPATCH DELAYS. IF SHARED POOL is stressed, VIOS to Dedicated Donating. VIOS to Dedicated. PLAN THE VIOS CORES IN THE SALES ALL PERF. FRAME CAN BE IMPACTED P7 PERF: SPEED TO CRUNCH DATA No Data No Speed!! Hi! VIO Server No I/Os, No network during 4 ms!! BAD FOR PERF I m back! VIO Server SELL/BUY THE CORES FOR THE VIOSERVERS 1) SHARED WITH CE/VP > ) DED. DONATING/DEDICATED Share Pool Core 0 Core 1 Core 2 Core 3 Core 4 Core 5 vp10 vp11 vp12 vp20 vp20 vp20 Shared Pool CO SHAR = 6 APP min= 0 vp20 vp10 vp11 vp12 vp20 vp10 vp11 vp12 vp20 vp10 vp11 vp12 vp20 11

12 PowerVM : Shared LPAR (CE, VP) The Choices physc (Physical Consumed in core unit) How many cores are you going to borrow in the Shared Pool? 1 Activity Profile of a LPAR This LPAR must become shared Time (day, week, month) 2 ways to answer physc Number of Virtual Processors CE:4.5 peak + margin CE:3.8 at peak* CE:2.8 under peak CE:0.6 minimum ❶ ❷ ENTITLEMENT COVERING CURRENT PRODUCTION NEED ENTITLEMENT NOT COVERING CURRENT PRODUCTION NEED 12 * Peak here is the required Production SLA = Production Need

13 Shared LPAR not covering Production Need # VP 2 1 physc 4 Prod. Need 3 Shared Capped LPAR CAP ON CE Shared Capped LPAR with a too low CE Do not cover production need. Flat top on curve or (MAX physc = CE) => Prod. is limited CE: 1.5 VP 3 4 Prod. Need physc Shared Uncapped LPAR CAP ON VP CE: 2.5 Shared Uncapped LPAR with too low VPs Do not cover production need. Flat top on curve or (MAX physc = VPs) => Prod. is limited We have free cores in the Share Pool (we re running on uncapped capacity) 13

14 Shared LPAR for Current Production Need # VP physc 4 Prod. Need Shared Capped LPAR MAX(physc) CAP ON CE CE: 3.8 Shared Capped LPAR with a CE at peak Cover Production Need. MAX (physc) = CE => Prod. Need is covered 2 1 physc 4 Prod. Need 3 Shared Uncapped LPAR # VP CAP ON VP CE: 3.8 You RESERVED the seats in the SHARE POOL your PRODUCTION NEEDS Shared Uncapped LPAR Does NOT Cover Production Need by itself! REQUIRES ENTITLEMENT DEFINED AT PEAK Little Margin Because this configuration offers a little margin, It is the most configured LPAR Type: Shared Uncapped. 14

15 Shared LPAR: Wrong Idea #1 physc #VP 4 4 Prod. Need CAP VP: 4 CE: 3.8 You RESERVED the seats in the SHARE POOL your PRODUCTION NEEDS Little Margin* Wrong Idea #1 If Little Margin is good, BIG MARGIN is BETTER! Wrong Idea. Thus, Bad Action: VPs are increased to the Long Term Sizing 15 A 0.95 B 0.38 CONFIGURATION A IMPACT ON CE/VP RATIO Data Demand Data Arrived! BAD FOR PERF B depends on Share Pool Capacity. B has now the dispatch delay B has lost its processor affinity B is using now more overhead. # VP 10 CAP 4 Prod. Need physc THIS SAME CURVE CONSUMES MORE OVERHEAD CONFIGURATION B Big Margin CE: 3.8 You reserved 3.8 seats It s what you NEED

16 Shared LPAR: Good Idea #1 physc #VP 4 4 Prod. Need CAP VP: 4 CE: 3.8 You RESERVED the seats in the SHARE POOL your PRODUCTION NEEDS Little Margin* Good Idea #1 Define your number of VPs for your Shared LPAR near the peak. Here it could have been: With 4VP CE/VP:0.98; With 5VP CE/VP:0.76; With 6VP CE/VP:0.63; Above 6VP, Remember Use the cores you bought! My 2-years planned growth: I need 10 VPs. DLPAR Add/Remove VP: 3.2s /2.4s vs a 2-years PERF impact. Plan your growth with Maximum VP value. PowerVM Hypervisor Dispatch Wheel = 10 ms vp12 vp12 vp12 vp12 vp12 vp12 vp11 vp11 vp11 vp11 vp11 vp11 vp10 vp10 vp10 vp10 vp10 vp10 Dispatch interval Potential uncapped Try to use the cores you bought! - Not to spend your time in dispatch wait. CONFIGURE WHAT YOUR PRODUCTION NEEDS NOW AS DESIRED VP DLPAR WILL TAKE CARE OF FUTURE GROWTH WITH MAXIMUM VP 16

17 DLPAR: Planned future growth with maximum VP Used at Activation to figure entitled capacity Used by DLPAR to check maximum allowed 17

18 Shared LPAR: Wrong Idea #2 VP Prod. Need physc C 0.62 D 0.25 CAP CONFIGURATION C - ACCEPTABLE IMPACT ON CE/VP RATIO 63% entitled to my needs % entitled to my needs The unused of the Share Pool * CE: 2.5 We are using the Unused at the optimum You reserved only 2.5 seats Wrong Idea #2 With less reserved CE, my shared pool has more Uncapped Capacity between LPARs # VP 4 CAP 4 Prod. Need physc Production depends on Shared Pool : With Conf. C, for 37% - With Conf.D for 74.5% 74.5% of Prod using now more overhead: sizing/capacity planning now differ 74.5% of Prod depend on Others LPARs: Prod is less predictable CE are decreased to have a LARGER SHARED POOL CAPACITY to configure more LPARs/VPs THIS SAME CURVE CONSUMES MUCH MORE OVERHEAD The unused of the Share Pool * CE: 1.0 You reserved only 1 seat CONFIGURATION D

19 Shared LPAR: Good Idea #2 # VP 4 Prod. Need physc CAP 3.8 at peak The unused of the Share Pool * CE: 2.5 You reserved only 2.5 seats POTENTIAL PB FOR FULL CAPACITY! BUT YOUR CE ENFORCES YOUR PRODUCTION SLA Good Idea #2 Define your CE at Your peak. Your Peak is your SLA, Your Peak can be below the curve peak. Here as your CE is now CE 2.5: With 4VP - Your CE/VP: > CE at 2.5 is acceptable with 4VPs Applying Rule, here: CE 3.8 (Peak Prod) -> VP=4 - CE/VP:0.98 CE 2.5 (SLA Prod) -> VP=3 - SLA CE/VP:0.83 It is recommended to have VP = ROUNDUP(CE) to the next whole or integer number Use WLE to size your CE 19

20 Making the right choice for the LPAR Type Easy to Manage : Are large share pools (with many LPARs) worth the work? Critical LPARs: Do they need to be at the mercy of others in the Shared Pool? The others could be a less controlled environment. Are they eligible to Dedicated Donating? Is their profile steady ones? Aren t they worth Dedicated? A pure controlled environment. Critical Appliances: VIO Servers Shared if really you want and their workload is acceptable. Be prepared to switch them to Dedicated Donating or Dedicated : size their cores as full dedicated. There is not a unique success receipt for Shared Pool. It s a multi-dimension configuration. Now, you ve some insights to configure quiet and balanced Shared Pools. 20

21 Real-Life Example 32 cores cores 21 This frame is an IDEAS Sizing result : Production 1. We need still to configure Production 1 + Disaster Recovery Prod 2 VIOs are corporate sized forgotten One Number Sizing effect. What s our problems?

22 Real-Life Example Some sizing problems Production 1 32 cores Planned Result direct from One Number sizing (Production 1 + Production 1 + Disaster Recovery) 6.4 unplanned 25.6 cores 22

23 Real-Life Example Solution CE/VP = 0.1 (illusion) Planned Result direct from One Number sizing (Production 1 + Production 1 + Disaster Recovery) 32 cores 6.4 unplanned 25.6 cores Production 1 and Production 1 are Rotating Production : - At one instant, we will have only 90 VPs and only 9 CE VPs and nearly 9 CEs are going to be ceded With Disaster activation, we will have 180 VPs and 18 CEs for 32 cores. On Paper /sizing spreadsheet, it works!!... BUT Ratio VP/Cores =196/32 = Who s eating the cake tonight? LPAR05 Prod 1: 9 LPAR05(2) = 7 + Dis. (9)=16 cores from > LPAR05 max VPs = 9.6 vs 20 VPs actives. 23

24 Real-Life Example trying to fix Sizing/Configuration It improves the ratio cores/vp Increase automatically CE/VP But did we fix Performance? Experience Ratio cores/vp > 4 You re surely in trouble Ratio cores/vp < 4 You can be still in trouble Ratio cores/vp [2-2.5] You re surely good. High VP LPARs are sensitive. 24 Reduce VPs

25 Same Real-Life Example Fix Performance Testing their Web Application Server on Development/Integration Frame : Configured Web. AS LPAR : 20 VPs 3 CEs / Sizing: 2.4 cores Sizing does not give VPs and CE : that s the beginning of troubles Which Ratio CE/VP? Size a Box and you do not know the shoe size. Result : 1.1 core/jvm + 8 JVMs/LPAR -> Need : 8.8 cores Java Recommendations and Tuning : 0.7 core/jvm (Gain 37%). CE : 3 Need : 5.6 consuming nearly 200% their entitlement. Tested ratio CE/VP is : 0.15 (=3/20) for a need of CE/VP 0.28 (5.6/20). 25

26 Same Real-Life Example Fix Performance I turned the same test in dedicated with 6 VPs (due to physc 5.6): what am I looking for? I m measuring Adventures in Uncapped Land for two Lands : Integration Frame Share Pool NOT EQUAL Production Frame Share Pool. My physc consumed is 0.4 core/jvm (3.2 core/lpar) Sizing with Updated Java stack Testing with obsolete Java bit. Sizing is 0.3/JVM I have 0.4/JVM Difference is 25% - Acceptable rnargin for obsolescence. Box Sizing : 2.4 cores Measure 8 JVMs : 3.2 cores That s why IBM can only publish Dedicated Performance Metrics. Dedicated CE/VP=1 / Configured CE/VP=0.1 : Larger difference with our Metrics. Entitlement: 100% / Entitlement: 10% : 90% difference on Entitlement. 26

27 Real-Life Example End Of Story We defined a CE/VP profile per type of LPAR. Looking a frame, the CE/VP gives you the application it is behind. NEARLY FULL CAPACITY HALF CAPACITY PERFORMS 27

28 Production Non-Production Profiles Non_prod = Prod Size it if you configure it Non_prod = Half-Prod Size it if you configure it Non_prod = Min CE Prod Min CE 0.1/VP (1H2012) Min CE 0.05/VP (2H2012) Remain in Sizing Hands But Remain is no Sizing Addressing Thru only CE Approach a Sizing problem 28 In all cases, the share pool is in contention : 54 VP 32cores 54VP

29 FAQ 1 Your Attempt#1 to rescue the situation : 54 VP Priority 32cores 54 VP Non Priority Do I manage with LPAR WEIGHT which LPAR receive UNCAPPED capacity? NO : PowerVM has NO MECHANISM to DISTRIBUTE uncapped capacity. YES : It s A WEIGHT MECHANISM to ARBITRATE SHARE POOL CONTENTION. redproc.htm Documentation says: Uncapped weight is only used where there are more virtual processors ready to consume unused resources than there are physical processors in the shared processor pool. If no contention exists for processor resources, the virtual processors are immediately distributed across the logical partitions independent of their uncapped weights. This can result in situations where the uncapped weights of the logical partitions do not exactly reflect the amount of unused capacity. The weight only comes into play when the hypervisor has more uncapped lpars ready to run virtual processors that their are idle processors in the system. When the share pool is in contention, LPAR Performance is already decreased.. In all cases, the share pool is in contention : 54 VP Priority 32cores 54VP Non Priority 29

30 FAQ 2 LPAR A Weight is 5 LPAR B Weight is 10 : B consumes CPU 2x than A? NO : The probability that PartitionA gets the uncapped time is 5/(5+10) : 33 % The probability that PartitionB gets the uncapped time is 10/(5+10) : 66 % redproc.htm Documentation says: logical partition 3 receives two additional processing units for every additional processing unit that logical partition 2 receives. If logical partitions 2 and 3 both require additional processing capacity, and there is enough physical processor capacity to run both logical partitions, logical partition 2 and 3 receive an equal amount of unused capacity. In this situation, their uncapped weights are ignored. The weight only comes into play where the hypervisor has more uncapped ready to run virtual processors that their are idle processors in the system. In all cases, the share pool is in contention : 54 VP Priority 32cores 54VP Non Priority 30

31 FAQ 3 Your Attempt#2 to rescue the situation : 2 shared pools 54 VP 25 cores and 54 VP 7 Cores Multiple shared pools vs One helps the balancing of the uncapped capacity. Multiple shared pools : The other shared processor pools on these models can be configured with a maximum processing unit value and a reserved processing unit value. The maximum processing unit value limits the total number of processing unit that can be used by the logical partitions in the shared processor pool. The reserved processing unit value is the number of processing units that are reserved for the use of uncapped logical partitions within the shared processor pool. NO : Documentation says: All three logical partitions compete for the same unused physical processor capacity in the server, even though they belong to different shared processor pools. The uncapped capacity is coming from the whole frame, independently of the multiple share pools. 54 VP 7 Cores can go after the idle capacity of 54 VP 25cores Thus, Maximum Processing Unit Value : 7 cores 54 VP 25 cores can go after the ilde capacity of 54VP 7 cores Thus, Maximum Processing Unit Value : 32 cores 31

32 FAQ 4 Why doesn t VP Folding take care of the issue of the CE/VP ratio being too low (i.e., < 0.6)? A very good question. As VP Folding is folding Virtual Processors, we could expect that unnecessary Virtual Processors will be folded. Thus, increasing the implicit CE/VP ratio. But, this means that we are first saturating a core on its 4 threads with the workload : the VP is used at 100% of its capacity before we unfold a new VP. This is not the case at 1H2012. As explained in a previous slide, the default AIX SMT Approach is to unfold all Virtual Processors to ensure the workload will be spread across all the VPs on the primary thread of each core. If you oversize your number of VPs and your workload is fluctuating, you will generate a constant unfolding/folding movement. Everyone should be aware that to spread the workload on each core first gives single threaded performance, faster response time, and total LPAR level throughput. Default AIX Folding / SMT Approach Core 1 Core 2 Core 3 Core 1 Core 2 Core 1 Core N It is important not to oversize the number of VPs. Use DLPAR operations to increase their number 32

33 SUMMARY OF OUR RECOMMENDATIONS Do NOT Undersize Memory - THINK LONG TERM NOW Ensure all processors are fully populated with Local Memory. UPGRADE the OS Levels and The Software (new compilers) Number ONE! Effect -NO SIZING on a UNIQUE NUMBER (rperf and CPW) - NO BOX SIZING : VIO? Which VIO? SELL THE CORES for the VIO SERVERS: First) Shared with a ratio CE/VP >0.6 (min.) Then) Ded. Donating/Dedicated++ The Number of VP of each LPAR must be less or equal than the number of cores of the Shared Pool. The Maximum CE of an uncapped LPAR is defined by the Shared Pool Configuration, not its number of VPs (ratio cores/vp) Large UNCAPPED CAPACITIES make the Shared Pool LESS PREDICTABLE. The UNCAPPED Exposures are Processor Affinity Loss and Dispatch Delays. They appear in over-committed Shared Pool. Define your number of VPs for your Shared LPAR near your peak. Your Peak can be your SLA Configure what your PRODUCTION NEEDS NOW as DESIRED VP DLPAR will take care of Future Growth with MAXIMUM VP It is recommended to have VP = ROUNDUP(CE) to the next whole or integer number Make the Right Choice for your LPAR Type. 33 * Still, William Shakespeare

34 Performance tuning in a modern environment. Modern computing environments are complex with many different commands and interfaces to review performance: While the data is excellent it can be complex to determine whether the system is tuned for optimal performance or what can be done to improve 34 - IT Performance expertise does rely on Plumber-minded skills and -. the plumbing can be badly complex.

35 Introducing: Power Systems Performance Advisors Performance interpretation combined with effective visual cues to alert clients about the state of the system and opportunities to optimize OPTIMAL Current condition likely to deliver best performance INFORMATIVE: Context relevant data helpful in making adjustments WARNING: Current condition deviates from best practices. Opportunity likely exists for better performance. CRITICAL: Current condition likely causing negative performance impacts. 35

36 Easy to Use Power Systems Performance Advisors are low overhead stand alone applications that collect performance metrics, before analyzing the results to produce a report that summarizes the health of the environment and proposes potential actions that can be taken to address performance inhibitors. How does it work? STEP 1) Download Advisor Advisor STEP 2) Run Executable STEP 3) View XML File Partition Partition 36 Only a single executable is required to run to gather the data The VIOS Advisor, and Virtualization Advisor can be set to monitor for a period of time. The AIX Java Advisor results are available immediately. Open up.xml file using your favorite web-browser to get an easy to interpret report summarizing your system status. Where to download Suggestion / Comments? padvisor@us.ibm.com

37 Virtualization Advisor monitors shared processor usage, processor folding and memory virtualization AIX Java Performance Advisor checks Java and WebSphere settings, including Heap Size and Garbage Collection settings VIOS Performance Advisor monitors virtual storage throughput and performance, as well as advising on CPU capacity and memory use for the VIOS partition 37

38 Conclusions We have reviewed : Basic Concepts of PowerVM. Technical Insights on the balance on CE / VP for shared LPARs. We hope this will help you to size, configure and implement Power7 Servers. Don t size shoe boxes ; Size your NEEDS Performance is about PRODUCTION PRODUCTION NEEDS 38

39 Special notices This document was developed for IBM offerings in the United States as of the date of publication. IBM may not make these offerings available in other countries, and the information is subject to change without notice. Consult your local IBM business contact for information on the IBM offerings available in your area. Information in this document concerning non-ibm products was obtained from the suppliers of these products or other public sources. Questions on the capabilities of non-ibm products should be addressed to the suppliers of those products. IBM may have patents or pending patent applications covering subject matter in this document. The furnishing of this document does not give you any license to these patents. Send license inquires, in writing, to IBM Director of Licensing, IBM Corporation, New Castle Drive, Armonk, NY USA. All statements regarding IBM future direction and intent are subject to change or withdrawal without notice, and represent goals and objectives only. The information contained in this document has not been submitted to any formal IBM test and is provided "AS IS" with no warranties or guarantees either expressed or implied. All examples cited or described in this document are presented as illustrations of the manner in which some IBM products can be used and the results that may be achieved. Actual environmental costs and performance characteristics will vary depending on individual client configurations and conditions. IBM Global Financing offerings are provided through IBM Credit Corporation in the United States and other IBM subsidiaries and divisions worldwide to qualified commercial and government clients. Rates are based on a client's credit rating, financing terms, offering type, equipment type and options, and may vary by country. Other restrictions may apply. Rates and offerings are subject to change, extension or withdrawal without notice. IBM is not responsible for printing errors in this document that result in pricing or information inaccuracies. All prices shown are IBM's United States suggested list prices and are subject to change without notice; reseller prices may vary. IBM hardware products are manufactured from new parts, or new and serviceable used parts. Regardless, our warranty terms apply. Any performance data contained in this document was determined in a controlled environment. Actual results may vary significantly and are dependent on many factors including system hardware configuration and software design and configuration. Some measurements quoted in this document may have been made on development-level systems. There is no guarantee these measurements will be the same on generallyavailable systems. Some measurements quoted in this document may have been estimated through extrapolation. Users of this document should verify the applicable data for their specific environment. Revised September 26,

40 Special notices (cont.) IBM, the IBM logo, ibm.com AIX, AIX (logo), AIX 6 (logo), AS/400, Active Memory, BladeCenter, Blue Gene, CacheFlow, ClusterProven, DB2, ESCON, i5/os, i5/os (logo), IBM Business Partner (logo), IntelliStation, LoadLeveler, Lotus, Lotus Notes, Notes, Operating System/400, OS/400, PartnerLink, PartnerWorld, PowerPC, pseries, Rational, RISC System/6000, RS/6000, THINK, Tivoli, Tivoli (logo), Tivoli Management Environment, WebSphere, xseries, z/os, zseries, AIX 5L, Chiphopper, Chipkill, Cloudscape, DB2 Universal Database, DS4000, DS6000, DS8000, EnergyScale, Enterprise Workload Manager, General Purpose File System,, GPFS, HACMP, HACMP/6000, HASM, IBM Systems Director Active Energy Manager, iseries, Micro-Partitioning, POWER, PowerExecutive, PowerVM, PowerVM (logo), PowerHA, Power Architecture, Power Everywhere, Power Family, POWER Hypervisor, Power Systems, Power Systems (logo), Power Systems Software, Power Systems Software (logo), POWER2, POWER3, POWER4, POWER4+, POWER5, POWER5+, POWER6, POWER7, purescale, System i, System p, System p5, System Storage, System z, Tivoli Enterprise, TME 10, TurboCore, Workload Partitions Manager and X-Architecture are trademarks or registered trademarks of International Business Machines Corporation in the United States, other countries, or both. If these and other IBM trademarked terms are marked on their first occurrence in this information with a trademark symbol ( or ), these symbols indicate U.S. registered or common law trademarks owned by IBM at the time this information was published. Such trademarks may also be registered or common law trademarks in other countries. A current list of IBM trademarks is available on the Web at "Copyright and trademark information" at The Power Architecture and Power.org wordmarks and the Power and Power.org logos and related marks are trademarks and service marks licensed by Power.org. UNIX is a registered trademark of The Open Group in the United States, other countries or both. Linux is a registered trademark of Linus Torvalds in the United States, other countries or both. Microsoft, Windows and the Windows logo are registered trademarks of Microsoft Corporation in the United States, other countries or both. Intel, Itanium, Pentium are registered trademarks and Xeon is a trademark of Intel Corporation or its subsidiaries in the United States, other countries or both. AMD Opteron is a trademark of Advanced Micro Devices, Inc. Java and all Java-based trademarks and logos are trademarks of Sun Microsystems, Inc. in the United States, other countries or both. TPC-C and TPC-H are trademarks of the Transaction Performance Processing Council (TPPC). SPECint, SPECfp, SPECjbb, SPECweb, SPECjAppServer, SPEC OMP, SPECviewperf, SPECapc, SPEChpc, SPECjvm, SPECmail, SPECimap and SPECsfs are trademarks of the Standard Performance Evaluation Corp (SPEC). NetBench is a registered trademark of Ziff Davis Media in the United States, other countries or both. AltiVec is a trademark of Freescale Semiconductor, Inc. Cell Broadband Engine is a trademark of Sony Computer Entertainment Inc. InfiniBand, InfiniBand Trade Association and the InfiniBand design marks are trademarks and/or service marks of the InfiniBand Trade Association. Other company, product and service names may be trademarks or service marks of others. Revised February 9,

41 Notes on benchmarks and values The IBM benchmarks results shown herein were derived using particular, well configured, development-level and generally-available computer systems. Buyers should consult other sources of information to evaluate the performance of systems they are considering buying and should consider conducting application oriented testing. For additional information about the benchmarks, values and systems tested, contact your local IBM office or IBM authorized reseller or access the Web site of the benchmark consortium or benchmark vendor. IBM benchmark results can be found in the IBM Power Systems Performance Report at All performance measurements were made with AIX or AIX 5L operating systems unless otherwise indicated to have used Linux. For new and upgraded systems, AIX Version 4.3, AIX 5L or AIX 6 were used. All other systems used previous versions of AIX. The SPEC CPU2006, SPEC2000, LINPACK, and Technical Computing benchmarks were compiled using IBM's high performance C, C++, and FORTRAN compilers for AIX 5L and Linux. For new and upgraded systems, the latest versions of these compilers were used: XL C Enterprise Edition V7.0 for AIX, XL C/C++ Enterprise Edition V7.0 for AIX, XL FORTRAN Enterprise Edition V9.1 for AIX, XL C/C++ Advanced Edition V7.0 for Linux, and XL FORTRAN Advanced Edition V9.1 for Linux. The SPEC CPU95 (retired in 2000) tests used preprocessors, KAP 3.2 for FORTRAN and KAP/C from Kuck & Associates and VAST-2 v4.01x8 from Pacific-Sierra Research. The preprocessors were purchased separately from these vendors. Other software packages like IBM ESSL for AIX, MASS for AIX and Kazushige Goto s BLAS Library for Linux were also used in some benchmarks. For a definition/explanation of each benchmark and the full list of detailed results, visit the Web site of the benchmark consortium or benchmark vendor. TPC SPEC LINPACK Pro/E GPC VolanoMark STREAM SAP Oracle Applications PeopleSoft - To get information on PeopleSoft benchmarks, contact PeopleSoft directly Siebel Baan Fluent TOP500 Supercomputers Ideas International Storage Performance Council Revised March 12,

42 Notes on HPC benchmarks and values The IBM benchmarks results shown herein were derived using particular, well configured, development-level and generally-available computer systems. Buyers should consult other sources of information to evaluate the performance of systems they are considering buying and should consider conducting application oriented testing. For additional information about the benchmarks, values and systems tested, contact your local IBM office or IBM authorized reseller or access the Web site of the benchmark consortium or benchmark vendor. IBM benchmark results can be found in the IBM Power Systems Performance Report at All performance measurements were made with AIX or AIX 5L operating systems unless otherwise indicated to have used Linux. For new and upgraded systems, AIX Version 4.3 or AIX 5L were used. All other systems used previous versions of AIX. The SPEC CPU2000, LINPACK, and Technical Computing benchmarks were compiled using IBM's high performance C, C++, and FORTRAN compilers for AIX 5L and Linux. For new and upgraded systems, the latest versions of these compilers were used: XL C Enterprise Edition V7.0 for AIX, XL C/C++ Enterprise Edition V7.0 for AIX, XL FORTRAN Enterprise Edition V9.1 for AIX, XL C/C++ Advanced Edition V7.0 for Linux, and XL FORTRAN Advanced Edition V9.1 for Linux. The SPEC CPU95 (retired in 2000) tests used preprocessors, KAP 3.2 for FORTRAN and KAP/C from Kuck & Associates and VAST-2 v4.01x8 from Pacific-Sierra Research. The preprocessors were purchased separately from these vendors. Other software packages like IBM ESSL for AIX, MASS for AIX and Kazushige Goto s BLAS Library for Linux were also used in some benchmarks. For a definition/explanation of each benchmark and the full list of detailed results, visit the Web site of the benchmark consortium or benchmark vendor. SPEC LINPACK Pro/E GPC STREAM Fluent TOP500 Supercomputers AMBER FLUENT GAMESS GAUSSIAN ANSYS Click on the "Benchmarks" icon on the left hand side frame to expand. Click on "Benchmark Results in a Table" icon for benchmark results. ABAQUS ECLIPSE MM5 MSC.NASTRAN STAR-CD NAMD HMMER Revised March 12,

43 Notes on performance estimates rperf for AIX rperf (Relative Performance) is an estimate of commercial processing performance relative to other IBM UNIX systems. It is derived from an IBM analytical model which uses characteristics from IBM internal workloads, TPC and SPEC benchmarks. The rperf model is not intended to represent any specific public benchmark results and should not be reasonably used in that way. The model simulates some of the system operations such as CPU, cache and memory. However, the model does not simulate disk or network I/O operations. rperf estimates are calculated based on systems with the latest levels of AIX and other pertinent software at the time of system announcement. Actual performance will vary based on application and configuration specifics. The IBM eserver pseries 640 is the baseline reference system and has a value of 1.0. Although rperf may be used to approximate relative IBM UNIX commercial processing performance, actual system performance may vary and is dependent upon many factors including system hardware configuration and software design and configuration. Note that the rperf methodology used for the POWER6 systems is identical to that used for the POWER5 systems. Variations in incremental system performance may be observed in commercial workloads due to changes in the underlying system architecture. All performance estimates are provided "AS IS" and no warranties or guarantees are expressed or implied by IBM. Buyers should consult other sources of information, including system benchmarks, and application sizing guides to evaluate the performance of a system they are considering buying. For additional information about rperf, contact your local IBM office or IBM authorized reseller. ======================================================================== CPW for IBM i Commercial Processing Workload (CPW) is a relative measure of performance of processors running the IBM i operating system. Performance in customer environments may vary. The value is based on maximum configurations. More performance information is available in the Performance Capabilities Reference at: Revised April 2,

44 Copyright IBM Corporation