DaveWright
SolidFireFounder,VicePresidentandGM
March/April2016
AdditionalSolidFireResources
InFebruary2016NetAppcompleteditsacquisitionofSolidFire,
amarketleaderinallflashstoragesystemsbuiltforthenext
generationdatacenter.TechOnTapispleasedtowelcome
SolidFirefounder,DaveWright,tointroduceourreaderstothis
technology.DavestartedSolidFirein2010thethirdcompany
hefoundedtobuildauniqueflashstoragearchitecturethat
deliverstheperformance,automation,andscaletoadvance
thewaytheworldusesthecloud.
Videos,Blogs,andArticles
SolidFireandNetApp(Dave
Wright,video)
SolidFire+NetApp=Unstoppable
Force
SolidFire:ThePerfectAdditionto
NetAppsFlashPortfolio(Dave
Hitz,blog)
NetAppandSolidFire:Flashto
MeetAllYourDataManagement
Needs(LeeCaswell,interview)
TheBeginningoftheEndofthe
AllflashArrayMarket(Dave
Wright,blog)
Publiccloudarchitectureismakingitswayintoenterpriseand
serviceproviderdatacenters,creatinganewsetofchallenges
foradministrators.Tobesuccessfulwiththecloudmodelwith
dynamicallyallocatedpoolsofcompute,networking,and
storageinfrastructurehastobeextremelycosteffectiveand
deliverscalability,automation,andsupportformultitenancy
andmixedworkloads.Thesearethecoreprinciplesofthe
SolidFiredesign.
Podcasts
TheTechOnTapPodcastteamhasbeen
workinghardtobringyouthelateston
SolidFire:
Episode25:NetApphasAcquired
SolidFire(AndyGrimes)
Episode26:SolidFireDeepDive
(AaronDelp,AmyLewis,
JeramiahDooley,AndyBanta)
Episode27:Infrastructure
Automation(JoshAtwell)
Episode28:SolidFire
Virtualization(AaronPatten)
StorageFieldDayVideos
DaveWrightandtheSolidFireteam
presentedatStorageFieldDay5
including:
SolidFireSolutionOverview
ComparingModernAllFlash
Architectures
UltimateSoftwareChooses
SolidFire
UltimateSoftwareisaleadingproviderof
humancapitalmanagement(HCM)
solutionsdeliveredviaSoftwareasa
Service(SaaS).Morethan2,800
customersoperatingin150countries
countonUltimatesUltiProsuite.
Ultimatehastfourmaindatacentersin
NorthAmerica,builtonapod
architecturewith200500customersper
pod.Plaguedbystorageperformance
issues,UltimateturnedtoSolidFire,
originallypurchasingfiveSF3010nodes,
andquicklyadding66SF2405nodes,
withplanstopurchaseanadditional44
nodes.
Inadditiontoperformance,SolidFirewas
chosenforitssecurity,availability,and
integrationwithbothOpenStackand
VMware.PodssupportedbySolidFire
deliver4xtheperformance,whileQoS
eliminatesthenoisyneighborproblems
seeninotherpods.
ReadtheUltimateSoftwarecasestudy.
Ifyou'realongtimeNetAppuser,you'llbecurioustoknowmoreaboutSolidFire.Thisarticleexplainssomeof
thetechnologychoicesthatmakeitsdesigndifferentfromotherallflasharrays.
ScaleOut,SharedNothingArchitecture
MostallflashstoragesystemsincludingPureandXtremIOuseadualcontrollerdesigntoprotectagainst
failure.Twocontrollersshareaccesstoasetofdrives,andonecontrollertakesoverfortheotherintheeventof
afailuredataisprotectedwithsometypeofRAID.
TheSolidFiredesigntakesadifferentapproach,usingascaleout,sharednothingarchitecture.EachSolidFire
nodeisastandard1Ux86systemwith10internalMLCorTLCSSDs.Nodesareinterconnectedvia10GbE,and
nothingissharedbetweennodes.HostsaccessdataviaeitheriSCSIorFibreChannelblockprotocols.
ASolidFireclusterstartswith4nodesandcanscaleoutincrementallyto100nodes.Dataisautomatically
distributedacrossallnodesinacluster,soeachadditionalnodeexpandsperformanceandcapacitylinearly.A
storagevolumeisneverconstrainedbytheperformancelimitsofasinglecontroller.
Figure1)SolidFiredeliverslinearscaleoutofperformanceandcapacityinasharednothingdesign.
Source:SolidFire2016
Thisapproachhassignificantadvantages:
Nondisruptivescaleout/scalein.AddorremovenodeswithoutdisruptingserviceorcompromisingQualityof
Service(QoS).Dataisautomaticallyredistributedinthebackgroundacrossallnodes,maintainingbalanceasthe
systemgrows.
Instantresourceavailability.Newlyaddedstorageresourcesareinstantlyavailabletoeveryvolumewithinthe
system,eliminatingtheneedtoreallocatevolumesovernewdrives.
Abilitytomixnodes.Somescaleoutsystemsrequirenodestobeidentical.SolidFiregivesyoutheabilityto
mixnodesofdifferenttypesandgenerations.Aselectionofperformanceandcapacitypointsletsyouscaleto
matchyourneeds.
Simplercapacityplanning.Becausescaleoutoccursin1Uincrements,performanceandcapacitycanbe
addedinaverygranularfashion.Thiseliminatesrelianceonmultiyearcapacityandperformanceprojections.It
alsoeliminatesupfrontoverprovisioning,allowingyoutotakeadvantageofpricereductionsovertime.
Noforkliftupgrades.Newgenerationnodescansimplybeaddedtoanexistingcluster.Whenthetimecomes,
oldnodescanberemovedandretiredorrepurposed.Compatibilitybetweenstoragenodesisguaranteed.Each
timeyouaddanode,youareabletoaddthemostuptodatetechnology.
DataAssurance
Toprovidedataredundancy,SolidFiremaintainstwocopiesofeverydatablockontwoseparatenodesa
technologycalledHelixthatisbuiltintoourElementOSoperatingsoftware.Thisallowsaclustertosustain
applicationperformanceafterfailuresoccur.Italsotakesawaytheneedtohaveseparatestorageshelveswith
shareddriveaccess,makingthehardwarelesscomplexandlessexpensive.Forinstance,SolidFireusessingle
attachSSDsinsteadofmoreexpensiveenterprisegrade,dualattachSSDs.
Thesystemselfhealsquickly,reducingtheriskthatasecondfailurewilloccurbeforeredundancyisrestored.
Becauseaclusterrespondsgracefullytonodesgoingoffline,thefeaturealsofacilitatesnondisruptivehardware
andsoftwareupgrades.
SelfHealingfromFailures
Allresourcesinthesystemarealwaysintheactivepoolthereisnoneedtohavesparedrivesorsparenodes
sittingidleincaseoffailure.
Drivefailure.Ifadrivefails,thesystemautomaticallyrestoresfullredundancybyredistributingcopiesofdata
usingameshedrebuildprocess.Thereisnodegradedmodeoperationandnoperformancepenaltyduringa
rebuild.Theprocesstypicallycompletesin5minutesorless.Becauseofthespeedwithwhichfullredundancyis
restored,itprovidesalevelofdataprotectionthatexceedsRAID6inatypicalsystem.
Nodefailure.Becausedatacopiesaredistributedonseparatenodes,alldataremainsaccessibleifanodefails.
Connectionstothefailednodeareautomaticallyredirectedtoothernodes.Aswithadrivefailure,fullredundancy
isrestoredquicklyandautomaticallybymakingsuretherearetwocopiesofeachblock.
Nomatterthefailuremodedrive,node,backplane,networkfailure,softwarefailuretherecoveryprocessisthe
same.Becausetherecoveryworkloadisdistributedacrossallnodesinthecluster,redundancyisrestored
quickly,andnosinglenode(orapplicationworkload)takesaperformancehit.Themorenodesinthecluster,the
fastertheactivityoccursandthelowertheoverallimpact.
Figure2)Afteranodefailure,dataredundancyisrestoredbydistributingnewcopiesofallblocksfromthefailed
nodeacrossthesurvivingnodes.Performanceandcapacityutilizationincreaseevenlyacrossallnodes.
Source:SolidFire2016
GuaranteedPerformance
Tosupportmixedapplicationworkloadsandmultitenantenvironmentsonasinglecluster,SolidFireprovides
guaranteedQoS.UnlikeimplementationsthatprovideQoSonabesteffortbasis,SolidFireisabletoguarantee
performancetoeachworkload.
Youcanallocateperformanceandcapacityindependentlyforeveryvolumeinasystem.Whenyoucreatea
volume,yousimplysetthedesiredsize,andspecifythreeQoSparameters:Min,Max,andBurst.Ifyouchange
thesettingsonavolume,itwillimmediatelystartreceivingserviceatthenewlevels.
TheMinsettingdefinesaminimumlevelofperformancemeasuredinIOPS(weightedbyI/Osize).Thevolume
isguaranteedtodeliveratleastthatlevelofperformanceunderallcircumstances.TheMaxsettingdefineshow
manyIOPSavolumecanconsume.Becausehardratelimitscancreateproblemsforapplicationsatransient
VDIbootstormisagoodexamplethereisalsotheBurstparameter.Applicationsbuildupcreditswhenthey
runundertheirmaximumlimit,allowingthemtoburstforshortperiodswhennecessary.
Figure3)GuaranteedQoS.TheleftsideillustratestheeffectofnoisyneighborswithQoSdisabled.Afew
poorlybehavedworkloadsrobperformancefromeverythingelse.Ontheright,theeffectofenablingQoSwith
varioussettingsfordifferentworkloadscanbeseen.
Source:SolidFire2016
AstudyfromEnterpriseStrategyGroup(ESG)estimatesthatSolidFirewithguaranteedQoScaneliminateupto
93%oftraditionalstoragerelatedissuesincludingproblemscausedbyworkloadimbalances,monopolizationof
afixedsetofresources,insufficientresourcesinapool,movingVMs,inefficienttiering,andcontroller
bottlenecks.ThestudyconcludedthatguaranteedQoSandautomatedloadbalancingallowanorganizationto
consolidateagreatervarietyandvolumeofworkloadsonasinglestoragesystem.Withtraditionalstorageandno
QoS,youwouldeitherspendmoretimeaddressingperformanceissuesoroverprovisionstoragearraysto
minimizeproblems.
AutomatedManagement
TheSolidFiredesigneliminatesmuchofthecomplexitythatwouldotherwisecomplicateautomation.
Performanceandcapacityareglobalpools,andworkloadsareautomaticallydistributedacrossthecluster.
Provisioningisextremelysimple,andmanytraditionalstoragetasksareeliminatedsuchas:
Performancetuningandloadbalancing
Managingtiering,prioritization,orcaching
Shortstrokingoroverprovisioning
RAIDgroupandsparedrivemanagement
Generationalupgradesorplatformmigrations
EverythingSolidFiredoesisexposedthroughacomprehensiveRESTbasedAPI.Automationreducestheriskof
humanerrorassociatedwithcomplexadministrativetasks.
Figure4)SolidFireRESTAPI.TheSolidFireRESTAPIunderpinsallSolidFiremanagementinterfaces,plug
ins,andtools,andfacilitatescustomintegrations.
Source:SolidFire2016
TheSolidFireAPIenablesdeepintegrationwithmanagementandorchestrationplatformsandsupportsthe
developmentofuserfacingstoragecontrols.Italsoenablesrapiddeploymentofapplicationsandservices.Allof
SolidFirestoolsandallthirdpartyintegrationsincludingthoseforVMware,OpenStack,andothersarebuiltusing
theAPI.
ESGmakesthebenefitofSolidFireautomationclear,concludingthatSolidFirehelpsadministratorsspinup
virtualmachinesupto81%fasterandlowersoperatingexpensesupto67%versustraditionalstorage.
InlineDataEfficiency
SolidFireoffersavarietyofstorageefficiencytechnologies,includingglobalthinprovisioningandspaceefficient
snapshotsandclones.ThesearesimilarinprincipletoNetApptechnologiesyouareprobablyalreadyfamiliar
with.
Combinedwithmultilayercompressionandglobalinlinededuplication,thesetechnologiesincreasethe
effectivestoragecapacityofaSolidFirecluster.
EachSolidFirenodeincludesaPCIeNVRAMcardthatservesasawritecache.Whenahostwritesdata,the
writeisdividedinto4KBblocksthatareimmediatelycompressedandstoredinNVRAM.Eachcompressedblock
issynchronouslyreplicatedtoanadditionalstoragenode.Anacknowledgementisreturnedafterdatahasbeen
storedinNVRAMonbothnodes,sowritesareextremelyfastandperformanceispredictable.
Eachcompressedblockishashedusingasecurecryptohashalgorithm.Theresultingvalueservesasa
BlockID.TheBlockIDdeterminesblockplacement,resultinginacontentaddressedstoragesystemsimilarto
thoseusedinleadingobjectstores.Thehashalgorithmdistributesblocksacrossallnodesinarandomfashion
thatensuresanevendistributionofload.
BasedontheBlockID,theSolidFireDeduplicationBlockServiceidentifiesblocksthathavepreviouslybeen
written.Ifablockalreadyexists,metadataisupdatedaccordinglyandtheduplicateisdiscarded.
Thededuplicationprocessisinlineandglobaldeduplicationhappensacrosstheentirecluster,notpervolume
orpernode.
Thecombinationofinlinecompressionandglobaldeduplicationhassubstantialadvantages:
Reducesdrivewear:Repetitivewritesareeliminated,increasingSSDlife.
Increasessystemperformance:Systemresourceconsumptionisminimized.
Eliminateshotspots:Workloadsareevenlydistributedacrossentireclusters.
Theinlinecompressionalgorithmwaschosenbasedonspeed.SolidFirealsousespostprocesscompressionin
thebackground.Amorecomputationallyintensecompressionalgorithmfurtheroptimizesstoragecapacity
withoutimpactingperformance.
CompleteStorageCapabilities
ThisarticleprovidesafoundationtohelpyoubegintounderstandtheuniqueaspectsofSolidFiresallflash
scaleoutstorageplatform.Whilemanyofthekeypointshavebeentouchedon,youshouldknowthatSolidFire
offersacomprehensivesetofstorageservices:
Replication(synchronous/asynchronous)
Integratedcloudbackup
Snapshotsandclones
256bitencryptionatrest
Comprehensivelogging
Cloudbasedmonitoring
Securemultitenancy
Simultaneousmultiprotocolsupport(FC/iSCSI)
Deepintegrations:VMware,OpenStack,CloudStack
TofindoutmoreaboutSolidFire,checkouttheresourcelistinthesidebarincludedwiththisarticleorvisit
solidfire.com.
DaveWrightleftStanfordin1998tohelpstartGameSpyIndustries,whereheledateamthatcreatedabackend
infrastructurepoweringthousandsofgamesandmillionsofgamers.HelaterservedasChiefArchitectforIGN
afteritacquiredGameSpy.
In2007DavefoundedJungleDisk,apioneerincloudbasedstorageandbackup.Rackspaceacquiredjungle
Diskin2008,andDaveworkedcloselywiththeRackspaceClouddivisiontobuildacloudplatformthat
supportedtensofthousandsofcustomers.InDecember2009,DaveleftRackspacetostartSolidFire.
QuickLinks
TechOnTapCommunity
Archive