Upload
ca-technologies
View
33
Download
3
Embed Size (px)
Citation preview
World®’16
AdvancedDBA– PerformanceTuningforMUFPartIIKevinShumaVP,ProductManagementCATechnologies
MAINFRAMEANDWORKLOADAUTOMATION
MFX66EB
2 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
Abstract
ThisisatwopartsessionthatdescribesthevarioustechniquesusedtotuneaCADatacomenvironment(MUF)forthebestpossibleperformance.Includedwillbediscussionsonavailablemonitoringtools.
ThesessionwillcovervarioustechniquestoimproveperformancewithintheCADatacom®/DBenvironments.
VP,ProductManagement
CATechnologies
KevinShuma
3 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
Agenda
ITUNEDLASTYEARWHYDOITAGAIN?
UTILIZINGTHEIBMZIIPTOIMPROVETCO
SUMMARY
REDUCINGPHYSICALIOTOIMPROVETCO
NEWFUNCTIONALITYAVAILABLEINV15.0
OTHERCONSIDERATIONS
1
2
3
44
5
6
4 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
ITunedLastYearWhyDoItAgain?
§ Hardwarechanges– SystemzIntegratedInformationProcessor(zIIP)– Increasedrealmemory– NewerfasterDASD
§ Softwarechanges– Exploitationofthehardware– Increasedmemorycapabilities– Newfasterorlessresourceintensivefunctionality
HowoftendoIneedtotune?
5 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
Manysitesmainframesoftwarechargesarebasedongeneralpurposeprocessoruse§ Rememberthisgraph?
§ Typicallicensingmodelswouldchargebasedonthehighestrolling4houraverage– nottheoverallaverage– Highest4hour 65.93MSUs– Average4hour 20.14MSUs
7 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
UtilizingtheIBMzIIP
§ SystemzIntegratedInformationProcessor(zIIP)– Hardwarecostsonly– DoesnotaddtoMIPs/MSUs– doesnotaffectsoftwarelicenses– zIIPCPUsrunatfullspeed– no“reducedcapacity”CPUs– IBMlimittozIIPsperGeneralPurpose(GP)CPU
§ SoftwarevendorslikeCAcancodesoftwaretousezIIPaslongastheystaywithinagivensetofrules– Onlycertainkindsprocessing– cannotdispatchphysicalIO– Onlycodeyourown– cannotzIIPotherpartiescode
Availablespecialtyprocesseronmostzboxes
8 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
CADatacomutilizingtheIBMzIIP
§ CADatacomandzIIPSpecialtyEngine– zOSrelease12.0andabovewilloffloadMUFGPprocessingtozIIP– Release12.0– offloads~33%tozIIP– Release14.0– offloads~50%tozIIP– Release14.01– offloads~80%+tozIIP– Release15.0– offloads~90+tozIIP
Availablespecialtyprocesseronmostzboxes
9 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
WhatcouldzIIPinCADatacomgetme?AssumingthatMUFin25%oftheprocessingonyourbox
WithoutzIIPMax.MSU 65.93Avg.MSU 20.14
Withr12zIIP33%andMUF@25%Max.MSU 60.49Avg.MSU 18.50
Withr14zIIP50%andMUF@25%Max.MSU 57.69Avg.MSU 17.63
10 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
WhatcouldzIIPinCADatacomgetme?(cont’d)AssumingthatMUFin25%oftheprocessingonyourbox
WithoutzIIPMax.MSU 65.93Avg.MSU 20.14
Withr14.01zIIP85%andMUF@25%Max.MSU 51.92Avg.MSU 15.86
Withr15zIIP95%andMUF@25%Max.MSU 50.27Avg.MSU 15.86
Straightlinesavingsnottypical
11 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
CADatacomutilizingtheIBMzIIP
§ CustomeronV15.0– Ourlicenseisbasedona55MSUhighestrollingaverage– Beforer14,weexceededthe55MSU6outof12monthsincurring
significantupchargesonsoftware– Mostrecentcontract55MSUspermonth
§ Withaspecialratetobuy“extra55MSUs”fortheyear– Afterr14.01andnow15.0,
§ Wehaveonlyexceeded50MSUsonceduringaveryheavystartupmonth– Wewillnotusemanyofthe“extraMSUs”thisyear– Reducesourcostsonthemainframe
Availablespecialtyprocesseronmostzboxes
12 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
CADatacomutilizingtheIBMzIIP(cont’d)
§ Start-upoption– SMPTASK5,3,3,SRB
§ CurrentdefaultisTCB§ CodeSRBtorunSMPtasksasSRBs§ MaintaskrunsasTCB(aswellasothersubtasks)§ FirsttwoSMPtasksscheduledtorunmoreaggressively§ OtherSMPtasksscheduledsameaswheninTCBmode§ AllSMPtasksareeitherSRBorTCB
UtilizingzIIPintheMUF
13 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
CADatacomutilizingtheIBMzIIP(cont’d)
§ EachSRBSMPtaskpairedwithaDBSMAPRTCB– UsedforthingsanSRBcan’tdo– OrthingsazIIPcan’tdo– Passedbackandforthassynchronousrequests
UtilizingzIIPintheMUF
SRB-SMP1
SMA1
SRB-SMP2
SMA2
SRBs TCBs
14 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
CADatacomutilizingtheIBMzIIP(cont’d)
§ ExampleofSRB/TCBpairs
UtilizingzIIPintheMUF
15 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
CADatacomutilizingtheIBMzIIP(cont’d)
§ Customerexample:Largebank– weekofprocessing– RunningV14.01zIIPwithlargebufferpools
UtilizingzIIPintheMUF
16 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
CADatacomutilizingtheIBMzIIP(cont’d)
§ AnytechniquethatreducesIOreductioncanimproveTCObyswappingGPCPUforzIIP-ableCPU– Usingmorebuffers– UsingMRDFcoveredandvirtualdata– ReducingIO
§ Largerblocksizes§ Useofcompression(withzIIPpresent)§ Anythingelse
Finalthoughts
18 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
HowtoImproveTCOWithDatacom– ReduceIO
§ DatabaseenginesarephysicalIOconsumers
§ PhysicalIOsconsumesignificantamountsofCPU
§ ReductionofPhysicalIOleadstoreductionofCPUconsumed– LogicalIOs(memory)arecheaperinCPUthephysicalIO– LogicalIOsarezIIP-able(specialtyprocessoreligible)
§ Rememberthebankwith97%zIIP
TypicallyreducingIOwillsaveGPCPU
MaybeprocessedonzIIP
19 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
ReduceIO(cont’d)
§ Ensurethebufferpoolsarelargeenoughtoprovidehighlogicalreuse– IXX/DXXbuffersare64-bit
§ Increaseeachuntil5+reuse“fallsoff”(increaseinreasonableincrements)– DATA/DATA2buffersare31-bitor64-bit(V15)
§ Increaseeachuntil5+reuse“fallsoff”(increaseinreasonableincrements)
– Ifusing31-bitstorage(donotallow31-bittogetexhausted)– Monitorpagingtoensurepagingremainsreasonable
Standardbuffers
20 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
ReduceIO(cont’d)
§ Implementlargerblocksizeforselecteddataareas– Ifinusemakesuretheyarebeingusedefficiently
§ LargerblocksizecouldreduceIO– Increasefrom4Kto18KdoesnotsignificantlyaffectcostofIOretrieval– Themorerowsperblock,themoreopportunityforimprovedreuse– NeedtovalidateusingBVMs
§ NeedtobalanceDATA/DATA2bufferstoareausage– IncreaseDATA2asworkloadismovedtothelargesize– Ifusing31-bitstorage,donotallow31-bittogetexhausted
Goingtoalargerdatablocksize=buffersize
21 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
ReduceIO(cont’d)
§ Likeaprivatebufferpoolforaselectedarea– ProvidesamemorycachetoavoidIOwhenrequestedblockisnotina
buffer
UsingMRDFCOVERED
22 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
ReduceIO(cont’d)
§ FIRSTusestheleastamountofCPUtofindarequestedblock– Directmapping(datablock7isinmemoryslot7)– Doesnotadjusttoactivity– Maynotworkwellifareatobecoveredistoobig
§ ACTIVEusesmoreCPUtofindarequestedblock– Adjuststoactivity,similartobuffering– LRUisusedtodeterminewhichmemoryslotsarereused– Indirectmapping(datablock7maybeanywhereinthememory)– Workswellifthe“workingset”ofblocksfitswithintheamountof
Coveredmemory
UsingMRDFCOVERED
23 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
ReduceIO(cont’d)
§ Coveringisnota“set-onceandforget”process– Needsreviewingonamonthlybasis
§ Coveringallocationscanbechangedontheflytomatchprocessingneeds
§ Coveringshouldbetargetedtohighreadareas– IXXoverDATAifsamevalue
§ Canbepre-loadedduringnon-peakprocessing
UsingMRDFCOVERED
24 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
ReduceIO(cont’d)
§ Reviewstatsbelow– somegoodchoices,somebadchoices
UsingMRDFCOVERED
26 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
ReduceIO(cont’d)
§ Lookingforthebestareastocover
UsingMRDFCOVERED
27 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
ReduceIO(cont’d)ComparisonbuffersversusMRDFCOVERED
§ Buffersarethemostefficientuseofmemorywhensizedtomatchthedata,but– Databufferpoolsinv14arelimiteddueto31-bit
– BuffersarereusedbaseonLRU(LeastRecentlyUsed)
– Averyactivebatchreportagainstanon-essentialtablecouldflushalargeportionoftheactivebufferpool
§ COVEREDareas– Are64-bitbydefault– Coveredblocksarespecifictoagivenareaandarenotflushedbynonassociatedactivity,but§ COVEREDareasmuststillfunneltheirblocksbackthroughthebufferpools
§ Coveringlowactivityareasusesthesamespacethathighactivityareasuses
28 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
ReduceIO(cont’d)
§ Buildbufferpoolstoareasonablesizetosupporttypical“peakactivity”withinthesystem
§ UseCoveringtotargetkeyhigh-use(read)datasets
§ Alwaysmeasureeverychangetomakesureitpresentsbusinessvalue!!!
BestChoice– Usebothlargebufferpoolsandselectivecovering
29 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
ReduceIO(cont’d)
§ Implementcompression– Compressionreducestherowimagesomorerowsperblock– ReducesDASDutilization(reducescost)– Canimprovebufferreuse– CanimproveCOVEREDefficiency
§ DBcompressionrunsinMUFandiszIIPeligible– WithzIIPprocessors,thecompressionCPUcostsareshiftedtoless
costlyCPUdevices
UsingCompression
30 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
ReduceIO(cont’d)
§ Anoptionthatoncewasconsideredoutdated– DASDcheap– CPUExpensive
§ Maynowbebackinstyle– DASDischeapbutzIIPCheaper– AlwaysmeasurewithBVMs!!– Newcooloptionin14.02
§ Onlinecompress§ SeesessionMFX94S- ReducingTCOwithDatacomCompression(Presspack)
UsingCompression
32 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
NewfunctionalityavailableinV15
§ Databufferpoolsnowgoto64-bit– Reduced31-bitconsumption– Abilitytocreatelarger(ormore)bufferpools
§ CanbecombinedwithadditionalbufferpoolstobuildtailoredbufferpoolsthatresembleCOVEREDareas– Tailoredbufferpoolswouldhavesomeadvantageovercoveredareas– Norequirementtomove“through”abufferpoollikeCOVERED– BufferLRUprocessingishighlyefficient
§ Whichisbetter- verdictisstillout
Datapoolsnowin64bit
33 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
NewfunctionalityavailableinV15(cont’d)
§ Areyouseeinglotsof“otherSRB”withzIIP– ExampleMUFwithmediumbuffersandlotsofIOactivity(V14)
ImprovedzIIPutilizationwithreducedGPCPU
Cmd Jobname Jobnr Clocktime CPU-Time SRB-Time IOCount CPTime IIPtime IIPonCP IIPencl ___ SHUKE01 59656 00:10:33 2.473518 0.020703 1062 2.432321 0.041197 0.000037 0.041197 ___ @2DBRUP2 59696 00:00:53 0.134624 0.023081 109 0.134624 ___ QA14MUF4 59663 00:08:14 5.667642 1.258595 70677 3.145144 2.522497 0.000524 2.522497 ___ @3PWRUP2 59700 00:00:46 0.099999 0.017140 109 0.099999 ___ @1UNRED1 59686 00:01:06 0.387458 0.079477 109 0.387458 ___ @3PWRUP3 59701 00:00:51 0.130862 0.023287 109 0.130862 ___ @2DBRUP1 59695 00:00:53 0.372552 0.063932 109 0.372552 ___ @2DBRUP3 59697 00:00:53 0.138674 0.024207 109 0.138674 ___ @1UNRED2 59687 00:00:56 0.103296 0.019699 109 0.103296
TotalCPU 5.667642zIIPCPU 2.522497 45%BillableCPU 3.145145 55% GPCPU 1.886550 33%
OtherSRB 1.258595 22%
34 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
NewfunctionalityavailableinV15(cont’d)
§ Areyouseeinglotsof“otherSRB”withzIIP– UsingJARStocaptureotherSRB
§ zIIPSRBisnotreportedsinceitdoesnotrunonaGPCPU
ImprovedzIIPutilizationwithreducedGPCPU
OtherSRB
35 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
NewfunctionalityavailableinV15(cont’d)
§ V15reduces“otherSRB”withzIIP– ExampleMUFwithmediumbuffersandlotsofIOactivity(V15)
ImprovedzIIPutilizationwithreducedGPCPU
TotalCPU 5.581476zIIPCPU 4.248196 76%BillableCPU 1.333280 24% GPCPU 0.967602 17%
OtherSRB 0.365678 07%
Cmd Jobname Jobnr Clocktime CPU-Time SRB-Time IOCount CPTime IIPtime IIPonCP IIPencl ___ SHUKE01 59656 00:54:02 5.729934 0.053566 2102 5.240694 0.489240 0.000089 0.489240___ @2DBRUP2 60326 00:00:52 0.307367 0.051069 109 0.307367 ___ QA14MUF4 60316 00:01:10 5.581476 0.365678 152683 1.333339 4.248196 0.000026 4.248196___ @3PWRUP2 60323 00:01:05 0.264024 0.044803 109 0.264024___ @1UNRED1 60325 00:00:52 0.432173 0.066993 109 0.432173___ @3PWRUP3 60331 00:00:45 0.181963 0.028822 109 0.181963 ___ @2DBRUP1 60318 00:01:05 0.426242 0.072268 110 0.426242 ___ @2DBRUP3 60330 00:00:51 0.300624 0.045925 109 0.300624 ___ @1UNRED2 60317 00:01:10 0.532060 0.088708 109 0.532060
36 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
NewfunctionalityavailableinV15(cont’d)
§ V15reduces“otherSRB”withzIIP– ExampleMUFwithmediumbuffersandlotsofIOactivity(V15)
ImprovedzIIPutilizationwithreducedGPCPU
OtherSRB
37 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
NewfunctionalityavailableinV15(cont’d)NewdataprovidedinMUFEOJandALL_INFO_REPORT
Limitedversionofreduced“other-SRB”availableforusersinV14.02
39 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
Otherconsiderations
§ Otherconsiderations– AreyouwastingCPUcycles
§ Areyoucheckingout– Applicationerrors(DBRCsorSQLCODEs)– PXXdumpactivity– Accountingactivity– CBStempindexes– SQLtemptables– Sequentialreadahead
Thingsyouneedtocheck
40 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
Otherconsiderations
§ Otherconsiderations– Checkingyouractivitysnapshots
§ Lookfordifferent/increasingstatistics– CheckyourDBMSTLSTassembly
§ MakesureyouareprovingthemaximumbufferstoDBUTLTY§ Needtochangedatabuffersizesifyouchangeblocksizes
Thingsyouneedtocheck
41 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
SummaryAFewWordstoReview
Results– PartIandIIImprovingperformanceandreducingTCOcanbeaccomplishedinmanywayswithinCADatacomThetoptechniquesrevolvearound:
MoveCPUconsumptiontoinexpensivespecialtyprocessorsReducingIObyusingbufferingandcoveringtomaximizememoryuseReducingactivitiesthatconsumeCPUwithoutbenefitProgramsthatget“badreturncodes”asanormalmodeofoperationBuildingtemporarytablesorindexesEtc.
@CAWORLD#CAWORLD ©2016CA.AllRIGHTSRESERVED.42 @CAWORLD#CAWORLD ©2016CA.AllRIGHTSRESERVED.42
MainframeandWorkloadAutomation
FormoreinformationonMainframeandWorkloadAutomation,pleasevisit:http://cainc.to/9GQ2JI
43 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
RecommendedSessionsSESSION# TITLE DATE/TIME
MFX92S StrategyandVisionforCADatacomandCAIdeal 11/16/2016at12:45pm
MFX93S What’snewinCADatacom 11/16/2016at1:45pm
MFX94S ReducingTCOwithDatacomCompression(Presspack) 11/16/2016at3:00pm
MFX95SA ModernizingtheCADatacomdata – SQLAccessPartI 11/16/2016at3:45pm
MFX95SB ModernizingtheCADatacomdata – SQLAccessPartII 11/16/2016at4:30pm
MFX96S CADatacom/ADInstallation/UpgradeandMaintenance 11/17/2016at12:45pm
MFX97S CADatacom/ADReportUtilization,andinformationgatherforSupport 11/17/2016at1:45pm
MFX98SUsingSysview tomonitorandmeasureyourCADatacomenvironment– ADandDB
11/17/2016at3:00pm
MFX99S BirdsofaFeather/StumptheTechie! 11/17/2016at3:45pm
46 ©2016CA.ALLRIGHTSRESERVED.@CAWORLD#CAWORLD
©2016CA.Allrightsreserved.Alltrademarksreferencedhereinbelongtotheirrespectivecompanies.
Thecontentprovidedinthis CAWorld2016presentationisintendedforinformationalpurposesonlyanddoesnotformanytypeofwarranty. The informationprovidedbyaCApartnerand/orCAcustomerhasnotbeenreviewedforaccuracybyCA.
ForInformationalPurposesOnlyTermsofthisPresentation