Data management paper

Max data store from 1 billion, 2 billion,
and 3 billion elements, with each element
consisting of an average of 2 kilobytes.
The results demonstrate that even as the
volume of data increased at each each
memory conguration point, latency and
throughput remained stable. The latency
remained fairly constant at 4.2 to 4.3
milliseconds, while the TPS remained stable
between 45,230 and 46,915.7
The combination of BigMemory Max
running on the Intel Xeon processor E7 v2
platform easily handled the load generated
by the seven client machines producing
a maximum of 350 concurrent threads
over a 1 gigabit Ethernet network, even
when the server memory was scaled to
6 TB. The CPU utilization peaked at 15
percent, which means the server was not
being utilized to its maximum capacity. The
limiting factor in these tests was the 1
gigabit network and the number of client
machines. In a production environment
using 10 gigabit Ethernet to connect
clients and servers, a single server would
be capable of producing higher TPS and
handling a higher number of connections.
The capacity and throughput can be
driven even higher by combining multiple
1
For more information, please visit https://www.softwareag.com/corporate/images/SAG_Terracotta_CERN_RS_Jan14_WEB_tcm16-116352.pdf.
2
For more information, please visit http://terracotta.org/resources/success-stories/nancial-services/better-faster-fraud-detection-drives-higher-prots.
3
 Nocomputersystemcanprovideabsolutereliability,availability,orserviceability.RequirestheIntel
®
Xeon
®
processorE7-8800/4800/2800v2productfamiliesoranIntel
®
Itanium
®
9500series-basedsystem(orfollow-ongenerationsofeither).Built-in
reliabilityfeaturesavailableonselectIntel
®
processorsmayrequireadditionalsoftware,hardware,servicesand/oraninternetconnection.Resultsmayvarydependinguponconguration.Consultyoursystemmanufacturerformoredetails.
4
 PCIeLiveErrorRecoveryimplementationsarespecictoanOEM’simplementation.ConsultyourOEMforspecicrecoverycapabilities.
5
 Formoreinformationseethepaper“AccelerateBigDataAnalysiswithIntel
®
Technologies”availableatwww.intel.com.
6
 Withtheserverhardwaremaximumof6TBofRAM,theBigMemoryMaxdatastorewasconguredforamaximumof5.5TBinsteadof6TBinordertoleaveenoughRAMavailablefortheoperatingsystem.6TBofmemoryachievedusing
Micron*64GBload-reducedDIMMs(LR-DIMMs)builtwithMicron’slatest4Gbquad-die,stackedDRAMcomponents.Formoreinformation,seehttp://www.micron.com/products/dram-modules.
7
 BasedonIntelCorp.internaltesting.Thetestenvironmentconsistedofasingleserverandsevenclientmachines.TheserverwasconguredwithfouroftheIntel®Xeon®processorE7-4890v2productfamilyrunningat2.2Ghz,96x64GB
DDR3-1333LRDIMM(6TBtotal),2x10gigabitIntel®EthernetControllerX540-AT2,2xSeagate300GB15KRPMSASharddrivesinaRAID0conguration,andanLSISAS2308RAIDcontroller.TheoperatingsystemwasRedHatEnter-
priseLinux*6.4,andthesoftwareusedwasTerracottaBigMemoryMax*4.0.5,Java*1.7.0_25,JavaSERuntimeEnvironment(build1.7.0_25-b15),andJavaHotSpot*64-BitServerVM(build23.25-b01,mixedmode).Theclientmachineswere
conguredwithtwooftheIntelXeonprocessorE5-2697productfamilyrunningat2.7GHz,8x4GB1333MHzmemorymodules,a1TB7200RPMSATA6GB/s2.5”harddrive,andanIntel®EthernetControllerI350gigabitnetworkinterface
controller.TheoperatingsystemwasRedHatEnterpriseLinux6.4,andthesoftwareusedwasEhcache*2.8.0,Java1.7.0_25,JavaSERuntimeEnvironment(build1.7.0_25-b15),andJavaHotSpot64-BitServerVM(build23.25-b01,mixed
mode).TheclientsusedEhcache2.8.0toreadfromandwritetotheBigMemoryMax.Twobenchmarktestswererunagainsttheconguration.Thersttestmeasuredcomputescalingwherethenumberofthreadsoneachclientwasincreased
from1to128,withaxeddatasizeof500GB.Thesecondtestdemonstrateddatascalingwheretheclientthreadcountwasxedat50,whilethedatavolumewasincreasedtollthecapacityofeachmemoryconguration.
 INFORMATIONINTHISDOCUMENTISPROVIDEDINCONNECTIONWITHINTEL
®
PRODUCTS.NOLICENSE,EXPRESSORIMPLIED,BYESTOPPELOROTHERWISE,
TOANYINTELLECTUALPROPERTYRIGHTSISGRANTEDBYTHISDOCUMENT.EXCEPTASPROVIDEDININTEL’STERMSANDCONDITIONSOFSALEFORSUCH
PRODUCTS,INTELASSUMESNOLIABILITYWHATSOEVER,ANDINTELDISCLAIMSANYEXPRESSORIMPLIEDWARRANTY,RELATINGTOSALEAND/ORUSEOFINTEL
PRODUCTSINCLUDINGLIABILITYORWARRANTIESRELATINGTOFITNESSFORAPARTICULARPURPOSE,MERCHANTABILITY,ORINFRINGEMENTOFANYPATENT,
COPYRIGHTOROTHERINTELLECTUALPROPERTYRIGHT.UNLESSOTHERWISEAGREEDINWRITINGBYINTEL,THEINTELPRODUCTSARENOTDESIGNEDNOR
INTENDEDFORANYAPPLICATIONINWHICHTHEFAILUREOFTHEINTELPRODUCTCOULDCREATEASITUATIONWHEREPERSONALINJURYORDEATHMAYOCCUR.
 Intelmaymakechangestospecicationsandproductdescriptionsatanytime,withoutnotice.Designersmustnotrelyontheabsenceorcharacteristicsofanyfeaturesorinstructions
marked“reserved”or“undened.”Intelreservestheseforfuturedenitionandshallhavenoresponsibilitywhatsoeverforconictsorincompatibilitiesarisingfromfuturechangesto
them.Theinformationhereissubjecttochangewithoutnotice.Donotnalizeadesignwiththisinformation.
 Theproductsdescribedinthisdocumentmaycontaindesigndefectsorerrorsknownaserratawhichmaycausetheproducttodeviatefrompublishedspecications.Current
characterizederrataareavailableonrequest.ContactyourlocalIntelsalesofceoryourdistributortoobtainthelatestspecicationsandbeforeplacingyourproductorder.Copies
ofdocumentswhichhaveanordernumberandarereferencedinthisdocument,orotherIntelliterature,maybeobtainedbycalling1-800-548-4725,orbyvisitingIntel’sWebsite
at www.intel.com.
 Copyright©2014IntelCorporation.Allrightsreserved.Intel,theIntellogo,andXeonaretrademarksofIntelCorporationintheU.S.andothercountries.
*Othernamesandbrandsmaybeclaimedasthepropertyofothers. PrintedinUSA 0114/TA/PRW/PDF PleaseRecycle 330106-001US
servers into a Terracotta Server Array, as
BigMemory demonstrates linear scale out
of TPS at a constant low latency rate.
Systems architects can use this
predictability to harness the power of
BigMemory Max and the high performance
and large memory capabilities of the Intel
Xeon processor E7 v2 family to scale their
in-memory data management systems as
their needs require. As demand increases,
enterprises can simply add additional
compute resources to quickly scale their
systems. Architects can also use the
performance predictability of BigMemory
Max and the Intel Xeon processor E7
v2 family to design server arrays with
specic performance and latency goals.
Conclusion
By harnessing the power of Terracotta
BigMemory Max and servers equipped
with the Intel Xeon processor E7 v2
family, enterprises can build a foundation
that enables data analysis use cases
that were, up until recently, difcult
or impossible to implement. Intel and
Terracotta engineers demonstrated that
the combination of BigMemory Max and
the Intel Xeon processor E7 v2 family can
utilize up to 6 TB of memory on a single
server and deliver consistent, predictable
throughput and latency.
In-memory data management and analysis
is the next step in big data evolution.
BigMemory Max running on servers
equipped with the Intel Xeon processor E7
v2 family provides a path for enterprises
to scale up and out, which opens up new
avenues to reap the benets of real-time
big data analysis.
To learn more how Intel can accelerate your
big data solutions, visit www.intel.com/
bigdata.
For more information about Terracotta
BigMemory Max, visit
www.terracotta.org/products/
bigmemorymax.
Terracotta and Intel: Breaking Down Barriers to In-memory Big Data Management