翻譯以學(xué)號(hào)-姓名.原文和在同一文件中前

用提供自定義的應(yīng)用視圖，這可以去除程序和數(shù)據(jù)結(jié)構(gòu)的干擾；此外，該模型能夠維持?jǐn)?shù)據(jù)的完整性和一致性。根據(jù)這一HDIM模型，一種名叫UDMP的基于模式的系統(tǒng)被設(shè)計(jì)并實(shí)現(xiàn)。實(shí)驗(yàn)結(jié)果表明，本文模型和相應(yīng)的系統(tǒng)在表現(xiàn)物聯(lián)性方緒。“自頂向下”的垂直過(guò)程，因?yàn)橐淮尾樵?xún)操作不會(huì)影響到其他數(shù)據(jù)模式的存在。這種據(jù)的完整性和一致性的同時(shí)不破壞全局范圍內(nèi)數(shù)據(jù)資源的自治性。與數(shù)據(jù)查詢(xún)相比,那“黑盒”的實(shí)現(xiàn)機(jī)制不能使一個(gè)外部應(yīng)用接觸到系統(tǒng)的結(jié)構(gòu)和行為狀態(tài)，因此不能異數(shù)成的本點(diǎn)式實(shí)異據(jù)成（G()和全局本地視圖模式(G)V將標(biāo)式定為數(shù)據(jù)模之的視；V將數(shù)據(jù)源模式定義為在目標(biāo)模式之上的視圖；V了V和V同。V展技。典的V是TSIS。V具有較高的效率但可擴(kuò)展性差，這不適合物聯(lián)網(wǎng)的數(shù)據(jù)環(huán)境。V通過(guò)重寫(xiě)技術(shù)獲取數(shù)據(jù)。典型的V項(xiàng)目是信息集成。V能松應(yīng)底的態(tài)。，在V中將全局?jǐn)?shù)據(jù)查詢(xún)轉(zhuǎn)換為局部數(shù)據(jù)查詢(xún)的過(guò)程比V更加復(fù)雜，這將導(dǎo)致效率的降低。V著眼將前兩方的共特。式匹與生是式的個(gè)鍵點(diǎn)。模式匹配就是在某種特定的規(guī)則下尋找數(shù)據(jù)元間不同的方式。有研究[8]提出一利用器習(xí)找全視與數(shù)源之關(guān)的方。種法基全的則 [9]提出了一種基于數(shù)值的方法以生成相應(yīng)的。這種方法利用數(shù)據(jù)元素與模式的語(yǔ)義約束之間的關(guān)系建立。本第2對(duì)M第3部分設(shè)計(jì)與數(shù)據(jù)集成模本文提出了一種基于模式的異構(gòu)數(shù)據(jù)集成模型HDIM。首先，HDIM為全部底 的視圖，以便 其次，HDIM在視圖之上提供了一個(gè)合適的應(yīng)用視圖，這樣所有應(yīng)用都能根據(jù)各自定義1：鍵。鍵是構(gòu)成數(shù)據(jù)表的基本元素，它與實(shí)體數(shù)據(jù)源中的關(guān)鍵詞建立了直接關(guān)系。鍵可以表示成一個(gè)四元組：Ukey=(Ukname,Utname,{KMap},{Fk})。Ukname表示鍵的名稱(chēng)，Utname是鍵所屬的數(shù)據(jù)表的名稱(chēng)，{KMap}表示從鍵到數(shù)據(jù)源鍵集合的路徑。tname,kname)；Ukname是鍵的名稱(chēng)，UVname是視圖名稱(chēng)，tname是表到的原始表名，kname是鍵到的原始鍵名。一個(gè)鍵通過(guò)KMap與原始定義3：完整性關(guān)系。完整性關(guān)系定義為一個(gè)三元組：Fk={parentUkey,dependentUkeyF}；ParentUkey表示雙親鍵dependentUkey表示子鍵；這兩者都可表示為二元組：parentUkeydependentUkeyUKRefUtname,Ukname)Ukname表示統(tǒng)一鍵名,Utname是鍵所屬的數(shù)據(jù)表的名稱(chēng)。F是一個(gè)約束。當(dāng)雙親鍵的值鍵值發(fā)生變化，HDIM都會(huì)在全局范圍內(nèi)檢查鍵的完整性關(guān)系。只有完整性關(guān)系得到滿(mǎn)定義4：數(shù)據(jù)表。數(shù)據(jù)表是整個(gè)系統(tǒng)中邏輯上的、虛擬的表。它將所的變化。數(shù)據(jù)表定義如下：Utable=(Utname,{Ukey}).{Utable});；UVname是視圖的名稱(chēng)；Addr包含了獲取原始數(shù)據(jù)源信息的，Addr以對(duì)原始數(shù)據(jù)源的尋址與連接為目標(biāo)。Utable代表數(shù)據(jù)表。數(shù)據(jù)表中的輯數(shù)據(jù)模型，它有如下的形式：DataObjDOnameItems})；DOname表示數(shù)據(jù)對(duì)象名7：數(shù)據(jù)對(duì)象屬性。數(shù)據(jù)對(duì)象屬性是構(gòu)成應(yīng)用視圖的基本元素，可表示為一個(gè)三元組：Item=(Itemname,DOname,{IMap})。Itemname是屬性名稱(chēng)，DOname是數(shù)據(jù)鍵的路徑。一個(gè)對(duì)象屬性在對(duì)象中有唯一的名稱(chēng)，它能到多種數(shù)據(jù)源，不過(guò)相同的數(shù)據(jù)源路徑只有一個(gè)。定義8：對(duì)象屬性的路徑。IMap可表示為一個(gè)四元組：IMap=(Itemname,DOname,Utname,Ukname)；Iternname是屬性名稱(chēng)。DOname是對(duì)象名稱(chēng)，Utname是屬性到的數(shù)據(jù)表的名稱(chēng)，Ukname屬性到的數(shù)據(jù)鍵的名稱(chēng)。通過(guò)IMap，建立一個(gè)對(duì)象屬性到一個(gè)或幾個(gè)鍵的關(guān)系。UVs)；DOs是應(yīng)用數(shù)據(jù)集合的應(yīng)用數(shù)據(jù)視圖，UVs是應(yīng)用數(shù)據(jù)集合的視圖數(shù)據(jù)表和鍵確定了全局中分布式異構(gòu)數(shù)據(jù)源的表示。Ukey定義了數(shù)據(jù)的完整性與一致性關(guān)系。應(yīng)用程序提交數(shù)據(jù)對(duì)象DataObj作為數(shù)據(jù)請(qǐng)求。通過(guò)IMap和KMap，HDIM能夠?qū)崿F(xiàn)任務(wù)的，分解和解決。Ukey建立了完整性約束關(guān)通過(guò)視圖的方式，HDIM提高了數(shù)據(jù)集成的靈活性和可擴(kuò)展性。HDIM在中，存在三種視圖:應(yīng)用視圖,視圖和原始數(shù)據(jù)視圖；同樣在其中存在兩種關(guān)系:IMap,KMap。IMap將應(yīng)用視圖中的數(shù)據(jù)對(duì)象到視圖中的統(tǒng)一鍵上。通過(guò)異構(gòu)數(shù)據(jù)錄入的視圖，的數(shù)據(jù)定義了一種的表達(dá)式,KMap將視圖中的鍵到原始數(shù)據(jù)源中的鍵。的表格和的鍵為異構(gòu)數(shù)據(jù)源和分布式數(shù)據(jù)源建立一個(gè)全局的表達(dá)式。視圖隱藏了異構(gòu)數(shù)據(jù)源之間的制的數(shù)據(jù)服務(wù)。在過(guò)程中，DOname和項(xiàng)目指定應(yīng)用程序數(shù)據(jù)對(duì)象及其屬性名稱(chēng)，Utname和Ukname指定的密鑰屬于的鍵名和的表格，Addr實(shí)現(xiàn)在的視圖中的數(shù)據(jù)源位置，TNAMEKNAME代表分別對(duì)應(yīng)的原始數(shù)據(jù)源中鍵，即鍵到的鍵。圖1展示了一個(gè)數(shù)據(jù)對(duì)象與原始數(shù)據(jù)源中的表之間的2.1HDIMForwardFullLink:item×GV→KeyForwardUpperLink:item×GV→Ukeyletutname=Getutname(Obj,item,Imap)letukname=Obj×item×Imap→utnameGetukname=Getukname：Obj×item×Imap→uknameGetutname=Imap(Objname,itemname)GetUkey:utname×ukname×GV GetUkey(utn,ukn,GV)=letUks= letresults=findUKey(ukn,Uks)infindUKey:ukname×Ukey* →UkeyfindUKey(ukn,Uk(Ukn,Utn,km*,fk*))=MatchName(ukn,Ukn)→Uk,nullfindUKey(ukn,Uk:Uks)=findUKey(ukn,Uk):findUKey(ukn,Uks)=letr=findUkey(ukn,Uk)in;r=null→findUkey(ukn,Uks)MatchName:string×string→MatchName(s1,s2)s1=s2→ForwardLowerlink:Ukey×UV→lettname=Gettname(Utname,Ukname,Kmap)letkname=Gettname：Utname×Ukname×Kmap→tnameGetkname=Kmap-1(Utname,Ukname)Getkname：Utname×Ukname×Kmap→knameGettname=GetKey:tname×kname×UV GetKey(tn,kn,UV)=letks= letresults=findKey(kn,ks)inresultfindKey:kname× →findKey(kn,K(Kn,Tn))=MatchName(kn,Kn)→letr=findKey(kn,K)in;r=null→findKey(kn,10：任務(wù)。一個(gè)任務(wù)由一個(gè)四元組組成：GTaskop,AObj,W),op代表數(shù)據(jù)操作類(lèi)型，并且它的可能取值可以是："SELECT"(數(shù)據(jù)提取)，"DELETE"(數(shù)據(jù)刪除),"INSERT"(數(shù)據(jù)),"UPDATE"(數(shù)據(jù)更新)；A是數(shù)據(jù)屬性的集合，由對(duì)象屬性定義組成：A={attribute};attribute=(item,value)，itemvalue是數(shù)據(jù)值，當(dāng)op是UPDATE或INSERT時(shí),value值是對(duì)象屬性到的數(shù)據(jù)記錄值，當(dāng)op是SELECTDELETEvalue值為空；ObjObj={objname}；W(item,Obj,f,comp)，item表示對(duì)象屬性名稱(chēng)，Obj表示數(shù)據(jù)對(duì)象名稱(chēng)，f表示約束值，comp表示約束關(guān)系。GTask轉(zhuǎn)化成UTask或Task。選擇對(duì)象屬性的過(guò)程可以作如下形式化描述：GetItemFromTask：GTask×GV →{Item}letresult=FindItem(itemname,items,GV)inresultGetItemFromTask：GTask×GV→{Item}letresult=FindItem(itemname,items,GV)inresultFindItem:Itemname*×item*×GV →{item}GetItem(Itemname,items,GV)FindVkeys(Itemnames,items,GV)過(guò)程如圖2所示：2.2ImplTask：GTaskdataresultletutsks=IMapTasks(GT)in;letrtsks=RelateTask(utsks)in;letsubresults=ImplsubTask(tsks:rtsks)in;collectData(subresults);IMapTasks：GTaskUTask*IMapTasks(GT)=letuv=FilterSource1(GT)inletresult=CreateUTasks(GT,uv))in;FilterSource1:GTaskUVname*CreateUTasks：GTask×UVname* KMapTasks：UTaskKMapTasks(UT)=letuv=FilterSource2(UT)in;letresult=CreateUTasks(UT,uv))in;FilterSource2:UTaskkname*CreateTasks：UTask×kname* RelateTasks：UTaskTask*ImplsubTask:Task*dataset*ImplsubTask(tsk)=letSql=createSQLFromTask(tsk)in;letcon=createSQLConnection(tsk)in;letd=execuSQL(con,Sql)in;letsubresult=d(con,tinfo,d)每一個(gè)鍵都到物理數(shù)據(jù)源中的一個(gè)物理鍵。在這個(gè)虛擬數(shù)據(jù)庫(kù)中的表之間和鍵之間都存在約束關(guān)系，這些約束關(guān)系在視圖中定義了一系列完整性約束。當(dāng)一個(gè)物理數(shù)據(jù)中的鍵發(fā)生變化，其他全局范圍內(nèi)的，與之存在完整性或一致性關(guān)系的鍵的F}；ParentUkey代表雙親鍵，dependentUkey代表子鍵；這兩者都可表示為一個(gè)三元組：parentUkey=dependentUkey=UKRef=(Utname,Ukname)。Ukname代表鍵名稱(chēng)，Utname是鍵所屬的數(shù)據(jù)表的名稱(chēng)；F是一個(gè)約束。的約束關(guān)系Fk是邊。圖3是關(guān)系圖的一個(gè)樣例：2.3A在圖中，點(diǎn)A,B,......代表鍵，邊F1,F2,代表鍵之間的關(guān)系定義。在間的約束關(guān)系。關(guān)系樹(shù)定義如下：RTreeUK{node})；UK是根節(jié)點(diǎn)，UKUtname,Ukname),，{node}是節(jié)點(diǎn)集合.。在Rtree中，一個(gè)節(jié)點(diǎn)對(duì)應(yīng)一個(gè)鍵，邊對(duì)應(yīng)完整FknodeUKFk,{node})，{node是子節(jié)點(diǎn)的CheckPKs：Ukey*×GV×RTree→BooleanCheckPKs(Uk,GV,GTCheckPK(UkGV,GT)CheckPKs(Vks,GV,GT)CheckPK：Ukey×GV×RTree→BooleanCreatePTree:Ukey×GV→RTreeCreatePTree(Uk,GV)ukn’(Ukn,Utn,-,-)=Ukletnodes=CreateNodes(GetPFk(Uk,GV))in;RTree((ukn,utn),nodes);CreatePNodes：Ukey×Fk* node*CreatePNodes(Uk,fk)=CreatePNode(Uk,fk)CreatePNodes(Uk,fks)CreatePNode：Ukey×Fk×GVnode fk,GV)=letukref(utn,ukn)=GetPairKey(Uk,fk)in; in;letpnodes=CreatePNodes(uk,GetPFk(uk,GV))in;letresult=n(ukref,fk,pnodes)原型系UDMP的設(shè)計(jì)與實(shí)這一部分設(shè)計(jì)并實(shí)現(xiàn)了一種三視圖的雙UDMP（數(shù)據(jù)管理平臺(tái)。它UMDP為系統(tǒng)中的異構(gòu)數(shù)據(jù)對(duì)象創(chuàng)建了一個(gè)的全局邏輯視圖。通過(guò)這個(gè)的全局了定制的應(yīng)用視圖，這樣每個(gè)應(yīng)用都能根據(jù)自己的需要具體地定制數(shù)據(jù)對(duì)象來(lái)進(jìn)行CRUD（創(chuàng)建，恢復(fù)，更新，刪除）操作。平臺(tái)架構(gòu)如圖5所示：UDP由四個(gè)主要部分組成：管理，執(zhí)行引擎，應(yīng)用程序接口和數(shù)據(jù)庫(kù)接口理括三數(shù)視：數(shù)庫(kù)視圖圖應(yīng)用圖據(jù)SQLUDMP在一定實(shí)驗(yàn)與分4.1中間節(jié)點(diǎn)數(shù)據(jù)庫(kù)服務(wù)1.Suse器Enterprise2.forSuse 這一章在本章中，比較UDMP和hibernate數(shù)據(jù)平臺(tái)[10]的性能，hibernate是一個(gè)使用O/R和命令的持久性管理的主流數(shù)據(jù)中間件。并發(fā)線(xiàn)程對(duì)這些記錄做，查詢(xún)，修改和刪除操作。所用數(shù)據(jù)庫(kù)為Oracle數(shù)據(jù)庫(kù)。表 實(shí)驗(yàn)結(jié)120,0001k

同時(shí)100個(gè)并 數(shù)據(jù)持續(xù)300分

100條記

UDMPhibernate64.1理等其他方面壓力時(shí)，UDMP的效率比hibernate略低，但它仍然處在標(biāo)準(zhǔn)的范圍內(nèi)。本章選擇國(guó)家前兆平臺(tái)的應(yīng)用場(chǎng)景作為測(cè)試場(chǎng)景來(lái)進(jìn)行實(shí)驗(yàn)并評(píng)價(jià)UDMP的實(shí)際表現(xiàn)。試驗(yàn)結(jié)果表明，UDMP可以實(shí)現(xiàn)的數(shù)據(jù)和數(shù)據(jù)管理，并具有比Hibernate更好的表現(xiàn)。結(jié)需要地組織和管理。關(guān)于數(shù)據(jù)集成的先前研究中關(guān)注數(shù)據(jù)的查詢(xún)和建模，但關(guān)于環(huán)境中的異構(gòu)數(shù)據(jù)約束關(guān)系的研究是不充分的。為了解決這些問(wèn)題，本文提出了關(guān)于異構(gòu)的作整合HIMHIM模型構(gòu)建出應(yīng)用視圖——視圖——數(shù)據(jù)源視圖的層次結(jié)構(gòu)，其中，在應(yīng)用視圖中生成的相應(yīng)數(shù)據(jù)對(duì)象為應(yīng)用程序提供定制的數(shù)據(jù)服務(wù)；全局標(biāo)準(zhǔn)邏輯視圖在視圖中建立；通過(guò)以上三級(jí)對(duì)數(shù)據(jù)進(jìn)行加工。此外，在視圖中還在數(shù)據(jù)項(xiàng)之中建立了約束檢驗(yàn)?zāi)Ｐ停瑥亩鴺?gòu)建整個(gè)數(shù)的數(shù)據(jù)協(xié)作。在那之后，本文設(shè)計(jì)與實(shí)現(xiàn)了基于HDIM模型的原型系統(tǒng)UDP，同時(shí)考慮到了數(shù)據(jù)和業(yè)務(wù)在域的實(shí)際要求特點(diǎn)。為實(shí)現(xiàn)透明地?cái)?shù)源，UP置的正的據(jù)式數(shù)標(biāo)UDP也實(shí)現(xiàn)了應(yīng)用程序和數(shù)據(jù)結(jié)構(gòu)的分離，并對(duì)象化了上層應(yīng)用以提高執(zhí)行效率；UP為視圖提供了配置功能UDMP參考文LiyanLiu,HuaLi.MetadataModelinGridDatabaseanditsApplication.The9thInternationalConferenceoncomputerSupportedCooperativeWorkinDesign,pp.362-366.UK,(2005)GuoHao-Ming,MaShi-Long.Functionaldemand-drivenresourceaggregationmethodandimplementation.JournalofBeihangUniversity,2008,34(5),pp.260-262Yong-Qiang,MaShi-Long,JinWen.Methodfordataintegrationmechanismformaintainingdataintegrity.JournalofBeihangUniversity,2008,34(9),pp.1045-1048GuoHao-Ming,HaoGuo-Shun.HeterogeneousdataintegrationandcoordinationinNGG.JournalofBeihangUniversity,2008,34(2),pp.180-182WangGui-Lin,LiYu-Shun,etc.AserviceGridDynamicInformationAggregationModelandItsApplication.JournalofComputers2005,28(4),pp.541-548Ki-WonYeom,Ji-HyungPark.AnEvolutionaryApproachforDynamicReconfigurationinHeterogeneousDatabaseSchemas.IEEECongressonEvolutionaryComputationSheraton,Vancouver,Canada,2006.pp.162-166TienN.Nguyen.Object—OrientedSoftwareConfigurationManagement.ICSM’06.22ndIEEEInternationalConferenceonSoftwareMaintenance,2006,pp.35l354．M.Lenzerini.DataIntegration:ATheoertiealPesrpective[A],ProceedingsoftheACMSIGMOD-SIGACT-SIGARTsymposiumonPrinciplesofdatabasesystemsACMNewYork,NY,USA,2002,pp.233-HuChun-Ming,HuaiJin-Peng,WoTian-Yu,LeiLei..AnendtoendSupportServiceGridArchitecture.JournalofSoftware,2006,17(5),pp.1448-1458GavinKing,ChristianBaue.HibernateinAction[M]．ManningPublicationsAHeterogeneousDataIntegrationLiuHai1,1,LiuYunzhen1,WuQunhui1,Ma1Departmentofcomputerscienceandtechnology,Beihanguniversity,37xueyuan100083Beijing,.WiththerapiddevelopmentoftheInternetofThings(IOT),thedatamanagement,datamininganddataysisinIOTsystemsrequireimprovingtheusabilityofthemulti-sourced,distributed,autonomousandheterogeneousdatafromthesubsystems,makingtheaggregation,integrationandcollaborationofthedataafocusinresearch.AccordingtothecharacteristicsofbasicIOTdataenvironment,aHDIMisproposedbasedonthecomparisonandysisofthecurrentexistingdataintegrationapproaches.Thismodelcannotonlymaskthedataheterogeneityindistributions,butalsoprovidethecustomizedapplicationviewfortheupperapplications,whichcandecoupletheprogramsanddatastructures;additionally,themodelcanmaintaintheintegrityandconsistencyofthedata.BasedonthisHDIMmodel,apattern-map-basedsystemwiththenameofUDMPisdesignedandimplemented.TheexperimentsshowthattheproposedmodelandthecorrespondingsystemcanaddressthefeaturesoftheIOTwithrelativegoodperformance.Keywords:InternetOfThings;dataintegration;viewmapWiththerapiddevelopmentoftheInternetofThings(IOT),andthegrowingdemandsofdatamanagement,datamining,dataysisintheapplicationlayerofIOT,thebusinessdatainbothintra-industryandinter-industryneedtobeintegrated.Theintegrationofdatabasesinthevariousdistributedandheterogeneoussubsystemshas eahotresearchtopic.AsthetechnicalbasisofdataintegrationinIOT,Datamiddlewareisthekeytechniquetotheintegrationandmanagementofthedistributedandheterogeneousdata.Therefore,thestudyondatamiddlewarehasimportantsignificancefordataintegration.Forthecurrentpracticalapplications,alongwiththecontinuoussystemintegration,linkageandcollaborationbetweensubsystems,therearealargenumberofdifferenttypesofapplicationunitsintheapplicationlayer.Meanwhile,thestorageformatsanddataspecificationsarediverseinthedatalayer,whichconstituteanumberofheterogeneousdatasources.Theseheterogeneousdatasourcesmakethedatamanipulationanddataexchangemoredifficult.Thecurrentresearchworkstaketheimplementationof"Query"asthemainpurposeofdataintegration.Dataqueryisa"top-bottom"verticalprocess,becauseonequeryoperationdoesnotaffecttheexistenceofdatainanotherschema.Thisformofmodelconstructionandorganizationtakethedataasisolatedobjectsduringdatamanipulationandcannotestablishmutualrelationsbetweenthesedatainthepatterns.However,inthesystemsofIOT,thebasicdatasupportingenvironmentcallsforthemaintenanceoftheglobaldataintegrityandconsistency.Thisrequiresthedataintegrationplatformnotonlytoachievetheunifiedqueryofheterogeneousdata,butalsotomaintaintheintegrityandconsistencyoftheheterogeneousdatawithoutviolatingtheautonomyofthedataresourcesintheglobalscope.Comparedwithdataqueries,operationswhichmodifythedatawouldcausemorecomplexchecksfordataintegrityandthecorrespondingmodificationsfordataconsistency.Inthisprocess,itisrequirednotonlytorewriteand posethedataoperationtaskstofinishthecurrenttask,butalsotoestablishcorrespondingoperationaltasksinordertomaintainglobalintegrityandconsistency.Taketheseismicindustryasanexample,thedistributeddatasourcesinvariousregionsnationwideconstitutethebasicdatasupportingenvironmentforthesystem.Theorganizationandmanagementofthesedatasourcesmaybedifferent.ly,datamodelsneedstobeconstructedsoastobemappedintoeverydatasourcebasedonthebusinessneeds.Theintegrityandconsistencyconstraintsinthedatamodelshouldbeestablishedsoastochecktheglobaldataintegritytogetherwiththetriggermechanismofeachdatasource,therebyintegratingtheentiredatasourcestoensuringtheintegrity,consistencyandvalidityofdatainthebasicdatasupportingenvironment.Secondly,moreandmoreapplicationsforIOTsystemsarewrittenintheobject-way.Theseapplicationscannotseamlesslyworktogethertherationaldatabasesbecauseofthedifferencesintheirmodels,To ethisimpedancemismatchbetweentheobjectmodelandtherelationalmodel,anobject/relationalmapisessentialforthesystem.Thirdly,thedatasourcehaspoorscalability,whenanewdatasourceisadded,itisdifficulttofacilitatetheintegrationofthenewdataortodynamicallyadjusttothechangesofthedatasources.Thetraditionalmiddlewaremainlyadaptsthe"blackbox"mechanism.The"blackbox"implementationmechanismcannotmakeanexternalapplicationtoaccessthesystem'sinternalstructureandstatebehavior,andthuscannotbedynamicallyadjustedaccordingtotherelevantchangesoftheapplicationlayerorthedatasourcelayer.Thislimitsthesystem’sscalabilityanptabilitytothedynamicenvironment.Thepatternisthebasicstartingpointinheterogeneousdataintegrationapproach.Patternmapisthecoretechnologytoachieveintegrationofheterogeneousdatasources[1-7].Accordingtothedifferentmapstyles,thereareGlobal-As-Viewmode(GAV),Local-As-Viewmode(LAV)andGlobal-Local-As-Viewmode(GLAV).GAVdefinesthemodeasaviewabovethedatasourcemode;LAVdefinesthedatasourcemodeasaviewabovemode;GLAVcombinesthecharacteristicsofbothGAVandLAV.GAVinquiresdatathroughunfoldingtechnology.OneofthetypicalGAVprojectsisTSIMMIS.GAVperformshighefficiencybutpoorscalability,whichisunsuitableforIOTdataenvironment.LAVqueriesdatabytherewritingtechnology.ThetypicalLAVprojectistheInformationManifold.LAVcaneasilyadapttothedynamicchangesoftheunderlyingdata.However,theprocessoftransformingglobaldataqueryintoalocaldataqueryinLAVismuorecomplexthaninGAV,whichresultsinlowerefficiency.GLAVfocusesonthecharacteristicsofthosetwoformermethods[7].Thepatternmatchingandmapgenerationaretwokeypointsofpatternmap.Patternmatchingislookingfordifferentmodesofmapbetweendataelementsfollowingparticularrules.Study[8]proposesamethodusingmachinelearningtofindtherelationshipbetweentheglobalviewandthedatasources.Thismethodisbasedonaglobalschemarulesdirectoryautomaticallyfindsbetweenthesourceschemamapbetweenelements;Thismethodautomaticallyfindsrelationshipthebetweentheglobalviewandthedatasourceelementsbasedonarulesdirectory;Study[9]proposesavalue-basedmethodforgeneratingthecorrespondingmap.ThemethodestablishesmapbyutilizingtherelationshipbetweentheelementsandsemanticconstraintsofTherestpartofthepaperisconstructedasfollows:Section2givesadetailedinstructionofHDIM.Section3providesdetaileddesignandimplementationoforiginaldatamiddlewaresystemUDMP.Section4conductsexperimentsandevaluationofUDMP.AndSection5concludesthepaper.DataIntegrationModelDefinitionofThispaperpresentsaHeterogeneousDataIntegrationModel(HDIM)basedonpatternmap.,HDIMcreatesagloballyunifiedviewforallunderlyingdatasourcessoastoshieldtheheterogeneousdatasourcesanddatastructuresthroughmap.Second,HDIMprovidesaappropriateapplicationviewabovetheunifiedviewsothatallapplicationscanbecustomizedaccordingtotheirneeds.Definition1.Unifiedkey.AunifiedkeyisthebasicelementtoconstituteaUnifieddatatable,itestablishesadirectmaprelationshipwithKeysintheentitydatasource.Unifiedkeyconsistsoffour-tupleform:Ukey=(Ukname,Utname,{KMap},{Fk});Uknamerepresentsforunifiedkeyname,Utnameisthenameofunifieddatatablewhichunifiedkeybelongsto,{KMap}isthemappathfromunifiedkeytoasetofkeyindatasources;Definition2.Mappathsoftheunifiedkey.KMapconsistsoffour-tuple:KMap=(Ukname,UVname,tname,kname);Uknameisthenameoftheunifiedkey,UVnameisUnifiedviewname,tnameistheoriginaltablenamewhichunifiedtableismapto,knameistheoriginalkeynamewhichtheunifiedkeyismapto.AunifiedkeyestablishesmaprelationshiptotheoriginalkeybyDefinition3.Integrityrelationship.Integrityrelationshipisdefinedbytripleform:Fk{parentUkey,dependentUkey,F};ParentUkeyrepresentsforparentkey,dependentUkeyrepresentsforthesub-key;bothofthemaremadeoftupleform:parentUkey=dependentUkey=UKRef=(Utname,Ukname).Uknamerepresentsforunifiedkeyname,Utnameisthenameofunifieddatatablewhichtheunifiedkeybelongsto;Fisaconstrainedlabel.Whenaparentkeygeneratesvaluechanges,itwillaffectthevalueofsub-key.Thereareasetofintegrityrelationshipsineachunifiedkey.Foreachchangeofthekeyvalue,HDIMcheckstheintegrityrelationshipofthekeyinaglobalscope.Dhangescannotbecarriedoutunlesstheintegrityrelationsissatisfied.Definition4.TheUnifiedtable.Unifiedtableisthegloballyunifiedlogicalandvirtualtableinthesystem.Itdefinesalltheunderlyingdatasourcedatainstandardssothatthesystemcanadapttothechangesofdatabaseproductorthedatabasestructure.Theunifiedtableisdefinedasfollows:Utable=(Utname,{Ukey}).Definition5.TheUnifieddataview.TheUnifieddataviewconsistsoftripleform:UnifiedView=(UVname,Addr,{Utable});UVnameisthenameofunifiedview;Addrcontainstheaccessinformationtotheoriginaldatasource,Addraimsataddressingandconnectiontotheoriginaldatasource.Utablestandsfortheunifieddatatable.Aunifiedkeyintheunifiedtablemaymaptodifferentoriginalkeyintheunderlyingdatasource.Definition6.TheApplicationview.TheApplicationviewconsistsofapplicationdataobject.Dataobjectisthespecificlogicaldatamodeloftheapplicationitself,whichhasthefollowingform:DataObj=(DOname{Items});DOnamerepresentsdataobjectname,{Items}standsforacollectionofattributesofthedataobject.Definition7.Attributesofdataobject.Dataobjectattributesarethebasicelementsthatconstitutetheapplicationview,theyeachconsistsofatripleform:Item=(Itemname,DOname,{IMap});Itemnameistheattributename,DOnameisdataobjectname,{IMap}pathsetsfortheobjectattributemap;{IMap}isthemappathfromobjectattributetoasetofunifiedkeyinunifiedview.Oneobjectattributehasauniquenameintheobject,itcanmaptomultipledatasources,however,thesamedatasourcepathhasonlyoneDefinition8.Mappathsofobjectattribute.IMapconsistsoffour-tuple:IMap=(Itemname,DOname,Utname,Ukname);Iternnameistheattributename,DOnameisobjectname,Utnameistheunifiedtablenamewhichattributesmapto,Uknameistheunifiedkeynamewhichattributesmapto.OneobjectattributeunifiedkeyestablishesthemaprelationshiptooneorseveralunifiedkeybyIMap.Definition9.Dataresourceintegrationmodel.Thedataresourceintegrationmodelconsistsofatupleform:GV=(DOs,UVs);DOsistheapplicationdataviewfortheapplicationdataset,UVsistheunifiedviewofapplicationdataset.Unifiedtableandunifiedkeyestablishagloballyunifiedexpressionforheterogeneousanddistributeddatasources.Ukeydefinesintegrityandconsistencyrelationshipofdata.ApplicationssubmitteddataobjectsDataObjfordataaccessrequest.HDIMrealizesthetask positionandresolutionaccordingtoIMapandKMap.Ukeyestablishestheintegrityconstraintrelationshipinordertomaintaintheglobalintegrityofthedataincaseofdatachanges.Uviewmasksthedifferencesbetweenheterogeneousdatasources.HDIMimprovestheflexibilityandscalabilityofdataintegrationinaviewmapway.HDIMInHDIM,therearethreeviews:theapplicationview,theunifiedviewandtheoriginaldataview;italsohastwomaps:IMap,KMap.IMapmapsdataobjectsinapplicationviewtotheunifiedkeyintheunifiedview.Theuniformaccesstodatathroughaunifiedviewofheterogeneousdataentrydefinesaunifiedexpression,KMapmapsunifiedkeyintheunifiedviewtothekeyintheoriginaldatasource.Theunifiedtableandunifiedkeyestablishagloballyaunifiedexpressionforheterogeneousanddistributeddatasources.TheunifiedviewmasksthedifferencesbetweenheterogeneousdatasourcesandimprovesflexibilityandscalabilityofHDIM.Theapplicationsviewcanprovidecustomizeddataaccessserviceforspecificapplications.Duringthemapprocess,DOnameandItemspecifyapplicationdataobjectsandtheirattributesname,UtnameandUknamespecifyunifiedkeynameandunifiedtablewhichtheunifiedkeybelongsto,Addrrealizesthedatasourcelocationintheunifiedview,tnameandknamerepresentthecorrespondingkeyintheoriginaldatasourcerespectively,whichtheunifiedkeyismapto.Fig.1showsamapbetweenadataobjectandatableintheoriginaldatasource.Fig.1.MapofMapprocessfromdataobjecttooriginalkeycanbeformalizedasfollows:ForwardFullLink:item×GV→KeyForwardUpperLink:item×GV→letutname=Getutname(Obj,item,Imap)letukname=Obj×item×Imap→utnameGetukname=Getukname：Obj×item×Imap→uknameGetutname=Imap(Objname,itemname)GetUkey:utname×ukname×GV →UkeyGetUkey(utn,ukn,GV)=letUks= letresults=findUKey(ukn,Uks)inresultfindUKey:ukname×Ukey* →UkeyfindUKey(ukn,Uk(Ukn,Utn,km*,fk*))=MatchName(ukn,Ukn)→Uk,nullfindUKey(ukn,Uk:Uks)=findUKey(ukn,Uk):findUKey(ukn,Uks)=letr=findUkey(ukn,Uk)in;r=null→findUkey(ukn,Uks)MatchName:string×string→BooleanMatchName(s1,s2)=s1=s2→ForwardLowerlink:Ukey×UV→lettname=Gettname(Utname,Ukname,Kmap)letkname=Gettname：Utname×Ukname×Kmap→tnameGetkname=Kmap-1(Utname,Ukname)Getkname：Utname×Ukname×Kmap→knameGettname=Kmap-1(Utname,Ukname)GetKey:tname×kname×UV GetKey(tn,kn,UV)=letks= letresults=findKey(kn,ks)inresultfindKey:kname×Key* →KeyfindKey(kn,K(Kn,Tn))=MatchName(kn,Kn)→Key,nullfindKey(kn,K:Ks)=findKey(kn,K):findKey(kn,Ks)=letr=findKey(kn,K)in;r=null→findKey(kn,Ks)DataManipulationTasksofAdatamanipulationtasktakestheapplicationdataobje anentity,throughmap,tododataextractionormodificationoperationinoneormorecorrespondingoriginaldataDefinition10.Task.Ataskisconstitutedbyafour-tuple:GTask=(op,A,Obj,W),opstandsforthedataoperationtype,anditspossiblevaluecanbe:"SELECT"(dataextraction),"DELETE"(datadeletion),"INSERT"(adddata),"UPDATE"(dataupdate);Aisthecollectionofdataattributewhichisconstitutedbyobjectattributedefinition:A={attribute};attribute=(item,value),itemistheobjectattributename,valueisthedatavalue,whenopisUPDATEorINSERT,valueisdatarecordvalueswhichtheobjectattributeismapto,whenopisaSELECTorDELETE,valueisnull;Oisacollectionofdataobjects,Obj={objname};Wisasetofconstraintswhichconsistsoffour-tupleform:(item,Obj,f,comp),itemfortheobjectattributenames,Objforthedataobjectname,fistheconstraintvalue,compistheconstraintrelationship.Inthetaskexecutionprocess,lyselectallobjectspropertiesinthetask,andthentransformtheGTaskintoUTaskorTaskbymaptheseproperties.Theselectionofobjectpropertiescanbeformalizedasfollows:GetItemFromTask：GTask× →letresult=FindItem(itemname,items,GV)inresultGetItemFromTask：GTask×GV→{Item}letresult=FindItem(itemname,items,GV)inresultFindItem:Itemname*×item*×GV →{item}GetItem(Itemname,items,GV)FindVkeys(Itemnames,items,GV)HDIMconstructsthedatamodelandsubmitsthedataaccessrequestbasedonapplicationrequirements.Whenthetaskengineacceptsthisrequest,it posesforeachdatasourcetoperformasub-taskinaccordancewiththemapofthedataobject.Thesesub-tasksexecuteinthedistributedoriginaldatasource.Finalresultsofthesub-taskarecollectedandreturntotheapplication.DataaccessprocessisshowninFig.2:Fig.2.GTaskexecutionTaskexecutionprocesscanbeformalizedasfollows:ImplTask：GTaskdataresultletutsks=IMapTasks(GT)in;letrtsks=RelateTask(utsks)in;letsubresults=ImplsubTask(tsks:rtsks)in;collectData(subresults);IMapTasks：GTaskUTask*IMapTasks(GT)=letuv=FilterSource1(GT)inletresult=CreateUTasks(GT,uv))in;FilterSource1:GTaskUVname*CreateUTasks：GTask×UVname* KMapTasks：UTaskTask*KMapTasks(UT)=letuv=FilterSource2(UT)in;letresult=CreateUTasks(UT,uv))in;FilterSource2:UTaskkname*CreateTasks：UTask×kname* RelateTasks：UTaskTask*ImplsubTask:Task*dataset*ImplsubTask(tsk)=letSql=createSQLFromTask(tsk)in;letcon=createSQLConnection(tsk)in;letd=execuSQL(con,Sql)in;letsubresult=d(con,tinfo,d)Theentiredataenvironmentcanbeconsideredasavirtualdatabasethatconsistsofalargenumberofheterogeneousdatasources.Eachkeyofthisvirtualdatabaseismappedtoarealphysicalkeyinphysicaldatasources.Constraintrelationshipexistsbetweentablesandkeysinthisvirtualdatabase,whichformsaintegrityconstraintsintheUnifiedView.Whenaunifiedkeyinthephysicaldatagenerateschanges,thevalueofotherunifiedkeywillbeinfluenced,whichhastheintegrityorconsistencyrelationshipswithitin aglobalscope.Integrityrelationshipisdefinedbytripleform:Fk={parentUkey,dependentUkey,ParentUkeyrepresentsforparentkey,dependentUkeyrepresentsforthesub-key;bothofthemmadebytupleform:parentUkey=dependentUkey=UKRef=(Utname,Ukname).Uknamerepresentsforunifiedkeyname,Utnameisthenameofunifieddatatablewhichtheunifiedkeybelongsto;Fisaconstrainedlabel.IntheUnifiedViewallintegrityconstraintrelationsconstituteadiagram,wheretheunifiedkeysarethevertexandtheconstraintrelationshipFkbetweenunifiedkeysaretheedges.Fig.3isanexampleoftherelationshipdiagram:Fig.3.ConstraintrelationtreeofnodeInthefigure,PointA,B,......standforunifiedkey,whileEdgeF1, standforrelationshipdefinitionbetweenunifiedkeys.Inthedefinitionofintegrityconstraintrelationship,ifaunifiedkeyisasub-key,theintegrityrelationshipofthisunifiedkeyisconsideredasaninverserelationship,thentherelationshipoftheedgescorrespondingtothe keyiscalledbackwardedge.Onthecontrary,ifaunifiedkeyisaparent-key,theintegrityrelationshipofthisunifiedkeyisthenconsideredasanforwardrelationship,therelationshipoftheedgescorrespondingtotheunifiedkeyiscalledforwardedge.Duringtheexecutionofthetask,HDIMneedstotheselectthecorrespondingconstraintrelationshipsaccordingtooperatetypetocompletedataintegritychecks.Accordingtotheoperatetype,dataintegritycheckcanleadtothefollowingdifferentSelectoperation:sincetherearenodhanges,sodonotmakeanyintegritychecksintheglobalscope.Deleteoperation,Needstodoforwardintegritychecksinglobalscopeontheunifiedkeywhichisabouttochange.Insertoperation,Needstodobackwardintegritychecksinglobalscopeontheunifiedkeywhichisaboutt