K9WAG08U1A GPU 2KB 64B (Flash Memory) SLC 512 I/O 1 [1-2] 1 NAND 80μs 200μs 15ms Gray : [3] : (1) (2) (3) 5 6

Transcription

1 CHINESEJOURNAL OFCOMPUTERS Vol36 No8 Aug2013 : 1)2) 1) 1) 1) ( ) 2) ( ) PC TP311 DOI103724/SPJ Flash-BasedDatabase:StudiesTechniquesandForecasts WANGJiang-Tao 1)2) LAIWen-Yu 1) MENG Xiao-Feng 1) 1) (Schoolof InformationRenmin Universityof ChinaBeijing ) 2) (Schoolof ComputerScienceand TechnologyHuaiyin Normal UniversityHuaianJiangsu ) Abstract Withthegrowingpopularityofflash memoryflashhasbeenwidelyusedinvarious typesofapplicationssuchasmobiledevicespc machinesandserversflashasanewtypeof storagemediaunlikediskhasitsinherentpropertieswhichincludenon-volatilityhighaccess speedshockresistancelowerpowerandhighstoragedensityetcthesepropertieshowever wilbringnewchalengesfordatamanagementalthoughdisk-baseddatabasemanagementsys- tem hasthepowerfulabilityformanagingdataitscurrenttechniquescannotmakefuluseof highi/operformanceofflashmemoryifwetransferittoflashwithoutmodificationtherefore itisahottopictodesignandimplementflash-baseddatabasesystemsinthissurveythenew propertiesandtranslationlayerofflashisintroducedfirstlysecondlyseveraldatabasecritical techniquesonbuferindexqueryoptimizationandtransactionprocessingaresurveyedand classifiedthirdlywediscusssomeresearchissuesonflash-basedhybridstoragesystemsat lastbasedontheanalyzedproblemssuggestionsforfutureresearchworksareputforward Keywords flashmemorybuferindexqueryhybridstorage 1 CPU : : ( ) (2012AA AA013204) (11XNL010) 1978 (CCF) jiangtaow@ruceducn

2 K9WAG08U1A GPU 2KB 64B (Flash Memory) SLC 512 I/O 1 [1-2] 1 NAND 80μs 200μs 15ms Gray : [3] : (1) (2) (3) NOR (4) NAND NOR SLC 10 MLC 1 NAND (5) NAND GB 2% (SLC) (MLC) SLC 30% 1 MLC NAND 22 NAND 1GB [4] I/O /s (2KB) (2KB) (128KB) 127ms 137ms N/A 1 htp://wwwaldatasheetcom/pdf/177488/samsung

3 8 : (SolidStateDriveSSD) (Flash Transla- tionlayerftl) 31 I/O PostgreSQL849 TPC-C :CPU 2DuoCPU E830028GHz 2GB 400MB Ubuntu [5] 12 80GBIntelSSD 2 2 SSD /% 1 3 : (1) Alocator I/O (2) [46] Wear Leveler (3) Cleaner PostgreSQL 10 IBM Oracle indexhtml 3 4 htp://wwwfusioniocom/casestudies/myspace htp://wwwfacebookcom htp://www01ibmcom/software/data/soliddb htp://wwworaclecom/us/products/database/timesten/

4 ω P miss P dirty ω ω 1 P miss [7-8] ω 5 P miss P dirty I/O 2 3 :(1) I/O (2) (3) ACID (1)CFLRU CFLRU [9] CFLRU CPU CPU (CleanPage) (DirtyPage) p : 3 (1) p p ( ) (2) (2) p p (3) 3 CFLRU (3) q q ( ) p LRU p LRU (1) : Cost(p)=P miss C read+p miss P dirty C write (1) P miss p C read C write CFLRU p P dirty p (1) (2) CFLRU ( ) Cost(p)=P miss 1+P dirty Cwrite Cread =P miss (1+P dirty ω) (2) CFLRU LRU LRU ω [10-13]

5 8 : 1553 (2)FOR : (1) FOR [14] (2) FTL : Hotness(p)=α IOD p +(1-α) OR p (3) IOD p OR p p α [15] IOD p p 30% OR p p IOD p OR p FOR Hotness(p) C read /C write (1)FAB FAB [16] LRU FOR R cold R cold 4 SSD FAB 322 PDA FAB FAB 5 BPLRU [17] REF [18] 4 (2)CFDC CFDC [19] n P(c)= n-1 pi-pi-1 i=1 n 2 Age(c) (4) n pi-pi-1 Age(c) CFDC 3

6 CF-LRU [9] LRU FOR [14] FAB [16] CFDC [19] : : : : : : : : : AD-LRU [13] : 323 Flash-DBsim [20] LRU CF- FAB LRU [9] FAB [16] CFDC [19] LRU : 2KB 64 5(b) 20μs200μs 15ms TPC-C BenchmarkSQL PostgreSQL CFDC 713%287%CF-LRU 60%CFDC 40% 33 5 I/O 5(a) n TT T index T I/O gain(t) 5 LRU FAB gain(t)=scan(t)-scan(t index )-update(t index ) =α C read-β C write (5) α C read T T index α T I/O C read I/O update(t index ) β C write β I/O C write I/O C write C read B + -Tree B + -Tree

7 8 : UM-B + - tree 6 (1)LA-Tree Agrawal [21] LA-TreeLA-Tree LA-Tree LA-Tree (2)Lazy-UpdateB + -Tree [22] Lazy-UpdateB + -Tree B + -Tree B + -Tree ( ) {keyrecptrtype} key recptr 6 A B recptrnul type ( )Lazy-Update A B B + -Tree UM-B + -tree B + -Tree gain(g)=cost(r)+cost(r )-cost(r R ) R g R g t cost(r)cost(r ) 332 t g cost(r R ) t (1) B + -Tree μ -Tree B (3)UM-B + -Tree + -tree [23] B + -tree UM-B + -treeum-b + -tree μ-tree [24] 6 UM-B + -tree

8 μ-tree μ-tree 7 F A F MicroHash (AC F) 7 μ-tree [26] HashTreeHashTree Hash Hash Hash FD-Tree [27] HashTreeHash (3)PBFilter PBFilter [28] μ-tree USB PBFilter (Key μ -Tree Value) (2)MicroHash PBFilterkey Bloom Filter MicroHash [25] Hash Bloom Filter MicroHash Bloom Filter PBFilter 4 4 : LA-Tree [21] Tree : : Lazy-UpdateB + -Tree [22] B + -Tree : : μ-tree [24] B + -Tree : : MicroHash [25] Hash+Queue : ( ) PBFilter [28] Bloom Filter : : 34 [29] - GraceHash 4 I/O I/O 2 I/O I/O I/O :(1)

9 8 : 1557 (2) [31] I/O PostgreSQL CPU I/O CPU CPU 4 [29] I/O I/O Hash I/O [30] PostgreSQL PostgreSQL TPC-H TPC-H % (1)RARE-join Hash Ailamaki [32] PAX PAX Hash [33-34] PAX RARE- join RARE-join {id 1 id 2 id i } I/O id i i 341 I/O RARE-join PAX CPU (2)DigestJoin I/O PostgreSQL RARE-join - Li [35] DigestJoin ( 1) 4 DigestJoin :1 DigestDigest I/O Hash 2 I/O

10 DigestJoin Myers (3)Sub-Join [36] Sub-Join Sub-Join 5 Myers [37] - 5 RARE-Join [33] DigestJoin [35] PAX PAX - Sub-Join [36] Subset [37] - CUSTOMER MB 343 DigestJoin 16MB Hash DigestJoin [35] Sub-Join [36] Sub-Join 3 :Intel Corei5-2400@310GHz 8GB Ubuntu1210 TPC-H CUS- TOMER ORDERS ORDERS 15 2GB (1) 001% ~10% 8MB 8(a) 01% Digest- JoinSub-Join HashJoin DigestJoin Sub-Join Sub-Join SSD DigestJoin (2) 8(b) 3

11 8 : 1559 DigestJoin 35 HV-Logging LB-Logging [39] FlashLogging [40] FlashLogging USB FlashLogging USB 352 SLC FlagCommit [41] SLC FlagCommit ID 351 HV-recovery [38] HV-recovery HV-Logging HV-Logging {Tidelement FlagCommit 1977 PreAddressPreValue} Tid element PreAddress PreValue 6 6 HV-recovery [38] FlagCommit [41] SLC LB-Logging [39] FlashLogging [40] USB USB

12 (SSD) SSD SSD SSD (1) [44] : I/O (1)SSD I/O SSD SSD SSD SSD SSD (2) SSD I-CASH [45] SSD SSD (3) SSD I-CASH SSD 40% SSD (2) ( SSD ) SSD SSD :(1)SSD I/O Hystor [46] (2)SSD 41 SSD SSDSSD SSD SSD : (1) SSD (2)SSD 411 SSD SSD 412 SSD SSD Hash - [647] (3) SSD MZ-7PD SSD Grifin [42] SSD3 Grifin PostgreSQL SSD SSD TPC-C Bench- [43] SSD marksql (a) SSD PostgreSQL SSD (tpmc) 1 SSD SSD SSD I/O SSD 3 TPC-H 9(b)

13 8 : SSD : 8 1GB (1) SSD SSD 489% (2)SSD SSD SSD SSD SSD SSD 421 [48] IBM DB2 TAC TAC (32 ) SSD SSD TAC SSD [49] LC Clean-Write(CW) Dual-Write(DW) 9 SSD Lazy-Cleaning(LC)3 CW 42 SSD SSD DW SSD LC DW LC SSD 3 SSD :(1) LC LC SSD SSD I/O SSD (2) 10 CW DW LC LCFaCE [50] SSD FaCE SSD SSD FaCE :(1) SSD FaCE 10 SSD FaCE SSD SSD (2)FaCE SSD SSD

14 OracleExadata 1 QoS(QualityofService) SSD QoS SSD SSD QoSSSD hstorage-db [51] hstorage-db hstorage-db SSD I/O I/O SSD hstorage-db hstorage-db I/O OLAP I/O hstorage-db 7 7 SSD TAC [48] Both Write-through LC [49] Both LRU-2 Write-back FaCE [50] Both FIFO Write-back Exadata 1 Clean LRU Write-through : : DB2 : :SSD : : : SSD : I/O 422 T1 LRU -ThroughLRU SSD disksim40 2 SSDModel [52] C Ubuntu1210 4KB 4MB LRU -Through CW [49] LC [49] FaCE [50] LRU -Through Exadata CW LC [49] TPC-C BenchmarkSQL PostgreSQL 8 8 T %/177% T %/365% SSD T1T2 11(a) T1 4 FaCE FaCE SSD 11 T1T2 3 1 htp://wwworaclecom/us/products/database/exadata CW LC 2 htp://wwwpdlainuedu/disksim

15 8 : 1563 CW LRU -K T2 T1 11(b) LC FaCE (1) FaCE SSD SSD 20 SSD 5 (2) SSD SSD 51 (3) : (1) SSD CPU NoSQL 53 [53-54] (2) (Storage Class Memory SCM) (PhaseChange MemoryPCM) SCM PCM 52 PCM (3) 9 [56] 9 64B 08J/GB 12J/GB 20~50ns20~50ns PCM 64B 1J/GB 6J/GB 10 6 ~10 8 ~50ns ~1μs 4KB 15J/GB 175J/GB 10 4 ~10 5 ~25μs ~500μs 512B 65J/GB 65J/GB ~5ms ~5ms [55] PCM PC PCM PCM

16 (1)SSD PCM SSD PCM PCM SSD [57-58] PCM SSD PCM SSD PCM PCM SSD SSD (2)PCM PCM PCM [59] PCM- Logging PCMLogging PCMLogging PCMPCM SSD [1] ChangLPKuoT WEficientmanagementforlarge-scale flash memory storage with resource conservationieee PCM 54 6 TransactionsonStorage20051(4): [2] WuC HKuoT WAnadaptivetwo-levelmanagementfor theflashtranslationlayerinembeddedsystems//proceedings oftheinternationalconferenceon Computer Aided Design (ICCAD 06)SanJoseUSA2006: Research:PacificGroveUSA2007:1-12 nationalconferencebeijingchina2007:55-66 (Database Machine) VancouverCanada2008: I/O I/O USA1993: Oracle IBM [3] GrayJTapeisdeaddiskistapeflashisdiskRAMlocali- tyisking//proceedingsofthe GongShow Presentationat Third Biennial Conference on Innovative Data Systems ingsofthevldbendowment20092(2): [6] LeeS WMoonBParkCKimJMKimS WAcasefor flash memory SSD in enterprise database applications// ProceedingsoftheACM SIGMODInternationalConference [7] O NeilEJO NeilP EWeikum GTheLRU -K page replacementalgorithmfordatabasediskbufering//proceedings of ACM SIGMOD International Conference Washington [4] LeeS WMoonBDesignofflash-basedDBMS:Aninpage loggingapproach//proceedingsoftheacm SIGMODInter- [5] Canim MBhatacharjeeBMihailaG AetalAnobject placementadvisorfordb2usingsolidstatestorageproceed- [8] JiangSZhangX DLIRS:Aneficientlowinterreference recencysetreplacementpolicytoimprovebufercacheper- formance//proceedingsofthe ACM SIGMETRICS Confer- enceon Measurementsand ModelingofComputerSystems CaliforniaUSA2002:31-42

17 8 : 1565 [9] ParkSYJungDKangJKimJSCFLRU:Areplace- mentalgorithmforflash-memory//proceedingsoftheinter- nationalconferenceoncompilersarchitectureandsynthe- sisforembeddedsystemsseoulkorea2006: [10] JungHShim HParkSetalLRU -WSR:Integrationof LRUandwritessequencereorderingforflashmemoryIEEE TransactionsonConsumerElectronics200854(3): [11] JungHYoonKShim HetalLIRS-WSR:Integrationof LIRSand WritesSequenceReorderingforFlash Memory// ProceedingsofInternational Conference on Computational Scienceand Its Applications Kuala Lumpur Malaysia 2007: [12] LiZJin P QSu XetalCCF-LRU:A new bufer replacementalgorithmforflashmemoryieee Transactions onconsumerelectronics200955(3): [13] JinP QOu YH rdertlizad-lru:aneficient buferreplacementalgorithmforflash-baseddatabasesjournal ofdata & KnowledgeEngineering201272(2): [14] LvY FCuiBHeBSChen XOperation-awarebufer managementin flash-based systems//proceedings of the ACM SIGMODInternationalConferenceAthensGreece 2011:13-24 [15] ChenFKoufatyD AZhangX DUnderstandingintrinsic characteristicsandsystemimplicationsofflashmemorybased solidstatedrives//proceedingsofthe ACM SIGMETRICS InternationalConferenceon Measurementsand Modelingof ComputerSystemsSeatleUSA2009: [16] JoHKangJParkS YetalFAB:Flash-awarebufer managementpolicyforportablemediaplayersieee Trans- actionsonconsumerelectronics200652(2): [17] Kim HAhnSBPLRU:Abufermanagementschemefor improvingrandom writesinflashstorage//proceedingsofthe 7thUSENIX ConferenceonFileandStorageTechnologies SanJoseUSA2008: [18] SeoDShinD KRecentlyevictedfirstbuferreplacement policyforflashstoragedevicesieee TransactionsonCon- sumerelectronics200854(3): [19] OuYHarder TJin P QCFDC:A flash-awarebufer managementalgorithmfordatabasesystems//proceedingsof the5thinternationalworkshopondatamanagementonnew Hardware(DaMoN)RhodeIslandUSA2009: [20] JinPQSuXYueL HAflexiblesimulationenvironment forflashawarealgorithms//proceedingsofthe18th ACM Conference on Information and Knowledge Management HongKongChina2009: [21] AgrawalDGanesanDSitaramanR KetalLazyadap- tive-tree:an optimizedindexstructureforflash devices ProceedingsoftheVLDBEndowment20092(1): [22] OnSTHu H BLiYFlash-optimizedB + -TreeJournal ofcomputerscienceandtechnology201025(3): [23] XuCShouLDChenGetalUpdatemigration:Anefi- cientb-treeforflash-storage//proceedingsofthe15thinter- national Conference on Database Systems for Advanced ApplicationsTsukubaJapan2010:1-4 waresalzburgsalzburgaustria2007: [25] ZeinalipourD ZLinSKalogerakiVetalMicroHash: Aneficientindexstructureforflash-basedsensordevices// [24] KangD WJungDKangJUμ-Tree:Anorderedindex structurefornandflash memory//proceedingsofthe7th ACM & IEEEInternationalConferenceonEmbeddedSoft- Proceedingsofthe3thUSENIXConferenceonFileandStor- agetechnologiescaliforniausa2005:31-44 [26] CuiKJinP QYueL-HHashTree:Anewhybridindex forflash disks//proceedingsofthe12th Asia-Pacific Web ConferenceBusanKorea2010:45-51 [27] LiYHeBSLuoQYiKTreeindexingonflashdisks// Proceedingsofthe25thInternationalConferenceon Data EngineeringShanghaiChina2009: residentdatathroughpartitionedsummaries//proceedingof the17th ACM ConferenceonInformationand Knowledge ManagementCaliforniaUSA2008: [29] DoJPatelJMJoinprocessingforflashSSDs:Remembering pastlessons//proceedingsofthe5thinternational Workshop on Data Management on New Hardware Rhode-Island USA2009: [28] YinSYPucheralPMengX FPBFilter:Indexingflash- [30] BauschDPetrovIBuchmann AOntheperformanceof databasequery processing algorithms onflash solid state disks//proceedings of the 1st International Workshop ConferenceonITServiceManagementandItsSupportTou- lousefrance2011: [31] Bausch DPetrovIBuchmann A P Makingcost-based queryoptimizationasymmetry-aware//proceedingsofthe8th InternationalWorkshoponDataManagementonNew Hard- warescotsdaleusa2012:24-32 [32] AilamakiADeWitDJHilM DSkounakisMWeaving relationsforcache performance//proceedings ofthe 27th InternationalConferenceonVeryLargeDataBasesRoma Italy2001: [33] Shah M AHarizopoulosSWienerJLGraefe GFast scansandjoinsusingflashdrives//proceedingsofthe4th InternationalWorkshoponDataManagementonNew Hard- warevancouvercanada2008:17-24 [34] TsirogiannisDHarizopoulosSShah M AetalQuery processingtechniquesforsolidstatedrives//proceedingsof theacm SIGMODInternationalConferenceRhodeIsland USA2009:59-72 [35] LiYOnS TXuJLetalDigestJoin:Exploitingfast randomreadsforflash-basedjoins//proceedingsofthe10th InternationalConferenceon MobileData ManagementTai- peichina2009: [36] LiangZhi-ChaoZhouDaMengXiao-FengSub-Join:Que- ryoptimizationalgorithmforflash-baseddatabasejournalof FrontiersofComputerScienceandTechnology20104(5): (inChinese) ( Sub-Join: 20104(5): )

18 [37] Myers D On the use of NAND flash memoryin high performance relational databases [MS dissertation] Massachusets Institute of Technology Massachusets USA2008 [38] LuZe-PingMengXiao-FengZhou DaHV-Recovery:A ConferenceAthensGreece2011: higheficientrecoverytechniqueforflash-based database [50] Kang W HLeeS WMoonBFlash-basedextendedcache ChineseJournalofComputers201033(12): (in forhigherthroughputandfasterrecoveryproceedingsofthe Chinese) VLDBEndowment20125(11): ( HV-Recovery: [51] LuoTLeeRMesnierMetalhStorage-DB:Heteroge (12): ) neity-awaredata ManagementtoExploittheFulCapability [39] LuZPQiX YCao WMengXFLB-Logging:Ahighly of Hybrid Storage Systems Proceedings of the VLDB eficient recovery technique for flash-based database// Endowment20125(10): Proceedingsofthe13thInternationalConferenceofWeb-Age [52] AgrawalNPrabhakaranVWobbeTDesigntradeofsfor Information ManagementHarbinChina2012: SSDperformance//Proceedingsofthe2008USENIX Annual [40] ChenS MFlashLogging:Exploitingflashdevicesforsyn- TechnicalConferenceBerkeleyCAUSA2008:57-70 chronouslogging performance//proceedings of the ACM [53] UngureanuCDebnath BRago SAranya ATBF:A SIGMODConferenceRhodeIslandUSA2009:73-86 memory-eficientreplacementpolicyforflash-basedcaches// [41] OnSTXuJLChoiBetalFlagCommit:Supporting Proceedingsofthe29thInternationalConferenceonDataEn- eficienttransactionrecoveryinflash-based DBMSsIEEE gineeringbrisbaneaustralia2013: Transactionson Knowledgeand Data Engineering2012 [54] DebnathBKSenguptaSLiJSkimpyStash:RAM space 24(9): skimpykey-valuestoreonflash-basedstorage//proceedings [42] GokulSVijayanPMaheshBTed WExtendingSSDlife- oftheacm SIGMOD ConferenceAthensGreece2011: timeswithdisk-based writecaches//proceedingsofthe8th USENIXConferenceonFileandStorageTechnologiesSan [55] DoJZhangDFPatelJMDeWitDJFastpeak-to-peak JoseUSA2010: behavior with SSD bufer pool//proceedings ofthe 29th [43] YangP YJinP QYueL-HHybridstorage withdisk International Conference on Data EngineeringBrisbane based write cache//proceedings of the DASFAA2011 Australia2013: Workshopon Flash-based DatabaseSystemsHong Kong [56] ChenS MGibbonsPNathSRethinkingdatabasealgo- China2011: [44] rithmsforphasechange memory//proceedingsofthe5th Ioannis K Stratis V Flashing up the storage layer BiennialConferenceonInnovative DataSystemsResearch ProceedingsoftheVLDBEndowment20081(1): [45] AsilomarCaliforniaUSA2011:21-31 YangQRenJI-CASH:InteligentlycoupledarrayofSSD [57] LeeS W Moon BPark CetalAcceleratingin-page andhdd//proceedingsofthe17thinternationalconference on HighPerformanceComputerArchitectureSanAntonio loggingwithnon-volatile memoryieee Data Engineering TexasUSA2011: Buletin201033(4):41-47 [46] [58] ChenFDavidAZhangX DHystor:Makingthebestuse SunG YJooYChenYBetalAhybridsolid-statestor- ofsolidstatedrivesinhighperformancestoragesystems// agearchitecturefortheperformanceenergyconsumption Proceedingsofthe25thInternationalConferenceon Super ComputingTucsonUSA2011:22-32 [47] ChenS MGibbonsPBNathSPR-join:Anon-blocking joinachievinghigherearlyresultratewithstatisticalguaran- tees//proceedingsoftheacm SIGMODConferenceIndiana USA2010: [48] Canim MMihailaG ABhatacharjeeBetalSSDbufer- poolextensionsfordatabasesystemsproceedings ofthe VLDBEndowment20103(2): [49] DoJZhang D HPatelJetalTurbocharging DBMS buferpoolusingssds//proceedingsoftheacm SIGMOD andlifetimeimprovement//proceedingsoftheinternational Conferenceon High Performance Computer Architecture BangaloreIndia2010:1-12 mentglasgowscotlanduk2011: WANG Jiang-Taobornin 1978 PhDcandidateHiscurrentresearch interests include flash-based database systemshybridstoragesystems [59] GaoSXuJLHeB SetalPCMLogging:Reducing transactionloggingoverheadwithpcm//proceedingsofthe 20th Conference on Information and Knowledge Manage- LAIWen-Yubornin1989MScandidateHisinter- estsincludequeryprocessingandoptimizationinflash-based databasesystems MENG Xiao-Fengbornin1964PhDprofessor PhDsupervisorHisresearchinterestsincludewebdataman- agementcloud data managementmobiledata management socialcomputingflash-basedsystemsprivacy-preserving

19 8 : 1567 Background Flashmemoryemergesasanewelectronicdevicewith lowaccesslatencyno mechanicallatencylowenergycon- sumptionandhighresistancetoshock whencompared with magneticdisksnandflash memoryhasbeen widelyused asstoragealternativeforembedded systemsinthe past yearswiththeincreasingofcapacityanddecreasingofunit priceflash-basedsolidstatedisks(ssds)hasbeensteadily expandedintopersonalcomputerandenterpriseservermar- ketsflashmemoryisconsideredasthemainstoragedevice insteadofdiskinthenextgenerationtodaymanydiferent typesofflashdevicesarefindingtheirwayintothestorage hierarchyofdatabasemanagementsystems (DBMS)How- everflashdeviceisnotperfectsomeofitsphysicalproper- tiessuch as erase before-write and limited write-erase cycleswhichnotonlydegradethei/o eficiencyonflash memorybutalso weartheflashdatablocksoutquickly Traditionaldisk-baseddatabasecannotmakefuluseofhigh I/Operformanceofflash memoryifwetransferittoflash withoutmodificationconsequentlyitisnecessarytodesign andimplementflash-awaredatabasesystemswhichcanhan- dleflashchipconstraintselegantlyduringdatamanagements Inordertosolvetheaboveproblemsmoreandmoreat- tentionshavebeengiventothisarearecentyearsresearch- ersproposedseveralkeytechnologieswhichareneededtobe addressedonflash-baseddatabasesystemstherearealotof researchtopicsstilaheadwhichneedmorepeoplejumpon theflash-baseddatabaseresearchesthecontentofthispa- permainlyprovidesasummaryforrelated worksandhelps researchersdiscovertheinterestingissues Theworkwaspartialysupportedbythegrantsfromthe NationalNaturalScience FoundationofChina ( )theFundamentalResearchFundsfor thecentraluniversitiesandtheresearchfundsofrenmin UniversityofChina (11XNL010)theNationalHighTech- nologyresearchanddevelopmentprogram (863Program)of China(2012AA AA013204)