Distributed DBMS rau Enterprise

Lub CAP theorem yog lub hauv paus ntawm kev faib tshuab txoj kev xav. Tau kawg, qhov kev tsis sib haum xeeb nyob ib puag ncig nws tsis txo qis: cov ntsiab lus hauv nws tsis yog canonical, thiab tsis muaj pov thawj nruj ... Txawm li cas los xij, ruaj khov rau ntawm txoj haujlwm ntawm txhua hnub kev nkag siab β„’, peb nkag siab tias qhov theorem muaj tseeb.

Distributed DBMS rau Enterprise

Qhov tsuas yog qhov tsis pom tseeb yog lub ntsiab lus ntawm tsab ntawv "P". Thaum pawg pawg tau muab faib, nws txiav txim siab seb puas yuav tsis teb kom txog thaum pawg pawg tau mus txog, lossis muab cov ntaub ntawv uas muaj. Nyob ntawm cov txiaj ntsig ntawm qhov kev xaiv no, lub kaw lus raug cais raws li CP lossis AP. Cassandra, piv txwv li, tuaj yeem coj tus cwj pwm ntawm ob qho tib si, nyob ntawm tsis yog nyob ntawm cov chaw teeb tsa, tab sis ntawm qhov tsis muaj ntawm txhua qhov kev thov tshwj xeeb. Tab sis yog tias qhov system tsis yog "P" thiab nws tawg, yog li cas?

Cov lus teb rau lo lus nug no yog qhov xav tau me ntsis: ib pawg CA tsis tuaj yeem sib cais.
Dab tsi ntawm pawg no yog qhov tsis tuaj yeem sib cais?

Ib qho khoom tseem ceeb ntawm xws li ib pawg yog kev sib koom cov ntaub ntawv khaws cia. Feem ntau ntawm cov xwm txheej, qhov no txhais tau tias kev sib txuas ntawm SAN, uas txwv tsis pub siv CA cov kev daws teeb meem rau cov lag luam loj uas muaj peev xwm tswj tau SAN infrastructure. Txhawm rau kom ntau lub servers ua haujlwm nrog tib cov ntaub ntawv, yuav tsum muaj cov ntaub ntawv sib koom ua ke. Cov ntaub ntawv zoo li no muaj nyob rau hauv HPE (CFS), Veritas (VxCFS) thiab IBM (GPFS) cov ntaub ntawv.

Oracle RAC

Qhov kev xaiv tiag tiag hauv pawg thawj coj tau tshwm sim hauv 2001 nrog kev tso tawm ntawm Oracle 9i. Nyob rau hauv xws li ib pawg, ntau qhov chaw server ua haujlwm nrog tib lub database.
Oracle tuaj yeem ua haujlwm nrog ob qho tib si cov ntaub ntawv sib koom ua ke thiab nws cov kev daws teeb meem - ASM, Automatic Storage Management.

Txhua daim ntawv khaws cia nws tus kheej phau ntawv. Kev ua lag luam yog ua tiav thiab cog lus los ntawm ib qho piv txwv. Yog tias ib qho piv txwv tsis ua tiav, ib qho ntawm cov kab ke tseem muaj sia nyob (xws li) nyeem nws cov cav thiab rov kho cov ntaub ntawv ploj - yog li ua kom muaj nyob.

Txhua qhov xwm txheej tswj hwm lawv tus kheej cache, thiab tib nplooj ntawv (blocks) tuaj yeem nyob hauv caches ntawm ntau qhov xwm txheej tib lub sijhawm. Ntxiv mus, yog tias ib qho piv txwv xav tau nplooj ntawv thiab nws nyob hauv cache ntawm lwm qhov piv txwv, nws tuaj yeem tau txais los ntawm nws cov neeg nyob ze siv lub cache fusion mechanism es tsis txhob nyeem los ntawm disk.

Distributed DBMS rau Enterprise

Tab sis yuav ua li cas yog tias ib qho ntawm cov xwm txheej xav tau hloov cov ntaub ntawv?

Qhov peculiarity ntawm Oracle yog tias nws tsis muaj qhov tshwj xeeb xauv qhov kev pabcuam: yog tias tus neeg rau zaub mov xav kaw ib kab, ces cov ntaub ntawv kaw tau muab tso ncaj qha rau ntawm nplooj ntawv nco qhov chaw xauv kab nyob. Ua tsaug rau txoj hauv kev no, Oracle yog qhov kev ua tau zoo ntawm cov chaw khaws ntaub ntawv monolithic: qhov kev pabcuam xauv tsis tau dhau los ua lub raj mis. Tab sis nyob rau hauv ib pawg configuration, xws li ib tug architecture yuav ua rau khaus network tsheb thiab deadlocks.

Thaum cov ntaub ntawv kaw lawm, ib qho piv txwv ceeb toom tag nrho lwm qhov xwm txheej tias nplooj ntawv uas khaws cov ntaub ntawv muaj qhov tshwj xeeb tuav. Yog tias lwm qhov piv txwv xav tau hloov cov ntaub ntawv ntawm tib nplooj ntawv, nws yuav tsum tau tos kom txog thaum cov kev hloov pauv rau nplooj ntawv tau cog lus, uas yog, cov ntaub ntawv hloov pauv tau sau rau hauv phau ntawv teev npe ntawm disk (thiab kev lag luam tuaj yeem txuas ntxiv). Nws kuj tseem yuav tshwm sim tias nplooj ntawv yuav raug hloov pauv los ntawm ntau daim ntawv theej, thiab tom qab ntawd thaum sau nplooj ntawv rau disk koj yuav tsum paub seb leej twg khaws cov ntawv tam sim no ntawm nplooj ntawv no.

Random hloov kho tib nplooj ntawv hla ntau RAC nodes ua rau database kev ua tau zoo poob qis, mus rau qhov chaw uas pawg ua haujlwm tau qis dua li ntawm ib qho piv txwv.

Kev siv Oracle RAC kom raug yog muab faib cov ntaub ntawv hauv lub cev (piv txwv li, siv lub rooj sib koom ua ke) thiab nkag mus rau txhua pawg ntawm kev faib tawm los ntawm ib qho kev mob siab rau. Lub hom phiaj tseem ceeb ntawm RAC tsis yog kab rov tav scaling, tab sis kom ntseeg tau tias kev ua txhaum cai.

Yog hais tias ib lub node nres teb rau lub plawv dhia, ces cov node uas kuaj pom nws thawj zaug pib cov txheej txheem pov npav ntawm lub disk. Yog tias qhov ploj lawm tsis tau sau tseg ntawm no, ib qho ntawm cov nodes yuav siv lub luag haujlwm rau cov ntaub ntawv rov qab:

  • "frozes" tag nrho cov nplooj ntawv uas nyob rau hauv lub cache ntawm cov node uas ploj lawm;
  • nyeem cov cav (rov ua dua) ntawm cov node uas ploj lawm thiab rov ua dua cov kev hloov pauv uas tau sau tseg hauv cov cav no, ib txhij kuaj xyuas seb lwm cov nodes puas muaj cov nplooj ntawv hloov tshiab tshiab dua;
  • rolls rov qab tseem tshuav pauv.

Txhawm rau kom yooj yim hloov ntawm cov nodes, Oracle muaj lub tswv yim ntawm kev pabcuam - piv txwv virtual. Ib qho piv txwv tuaj yeem pab ntau yam kev pabcuam, thiab qhov kev pabcuam tuaj yeem txav mus los ntawm cov nodes. Ib daim ntawv thov piv txwv ua haujlwm rau qee qhov ntawm cov ntaub ntawv (piv txwv li, ib pab pawg neeg siv) ua haujlwm nrog ib qho kev pabcuam, thiab cov kev pabcuam rau lub luag haujlwm ntawm cov ntaub ntawv no txav mus rau lwm qhov thaum lub node ua tsis tiav.

IBM Pure Data Systems for Transactions

Ib qho kev daws teeb meem rau DBMS tau tshwm sim hauv Blue Giant portfolio hauv 2009. Ideologically, nws yog tus successor ntawm Parallel Sysplex pawg, ua los ntawm cov khoom siv "tsis tu ncua". Hauv xyoo 2009, DB2 pureScale tau tso tawm raws li software suite, thiab hauv 2012, IBM tau muab cov khoom siv hu ua Pure Data Systems for Transactions. Nws yuav tsum tsis txhob tsis meej pem nrog Pure Data Systems rau Analytics, uas yog tsis muaj dab tsi ntau tshaj li ib tug renamed Netezza.

Thaum xub thawj siab ib muag, lub pureScale architecture zoo ib yam li Oracle RAC: nyob rau hauv tib txoj kev, ob peb nodes yog txuas nrog ib tug ntau cov ntaub ntawv cia system, thiab txhua tus ntawm nws tus kheej DBMS piv txwv nrog nws tus kheej lub cim xeeb cheeb tsam thiab cov ntaub ntawv sib pauv. Tab sis, tsis zoo li Oracle, DB2 muaj kev pabcuam xauv tshwj xeeb uas sawv cev los ntawm cov txheej txheem db2LLM *. Hauv kev teeb tsa pawg, qhov kev pabcuam no tau muab tso rau ntawm qhov sib cais, uas yog hu ua coupling facility (CF) hauv Parallel Sysplex, thiab PowerHA hauv Cov Ntaub Ntawv Ntshiab.

PowerHA muab cov kev pabcuam hauv qab no:

  • tus neeg saib xyuas xauv;
  • ntiaj teb no tsis cache;
  • thaj tsam ntawm kev sib txuas lus sib txuas.

Txhawm rau hloov cov ntaub ntawv los ntawm PowerHA mus rau cov chaw khaws ntaub ntawv thiab rov qab, cov chaw taws teeb nco tau siv, yog li pawg sib txuas yuav tsum txhawb nqa RDMA raws tu qauv. PureScale tuaj yeem siv ob qho tib si Infiniband thiab RDMA hla Ethernet.

Distributed DBMS rau Enterprise

Yog hais tias ib tug node xav tau ib nplooj ntawv, thiab nplooj ntawv no tsis nyob rau hauv lub cache, lub node thov nplooj ntawv nyob rau hauv lub ntiaj teb no cache, thiab tsuas yog hais tias nws tsis nyob ntawd, nyeem nws los ntawm disk. Tsis zoo li Oracle, qhov kev thov tsuas yog mus rau PowerHA, thiab tsis yog rau cov neeg nyob sib ze.

Yog tias ib qho piv txwv yuav hloov ib kab, nws kaw nws hauv hom tshwj xeeb, thiab nplooj ntawv uas kab nyob hauv hom sib koom. Tag nrho cov xauv tau sau npe rau hauv lub ntiaj teb kev xauv tus tswj hwm. Thaum qhov kev sib pauv ua tiav, cov node xa cov lus mus rau tus tswj hwm xauv, uas luam cov nplooj ntawv hloov kho mus rau lub ntiaj teb cache, tso cov xauv, thiab invalidates nplooj ntawv hloov kho hauv lub caches ntawm lwm cov nodes.

Yog tias nplooj ntawv nyob rau hauv cov kab hloov kho tau raug kaw lawm, ces tus thawj tswj lub xauv yuav nyeem cov nplooj ntawv hloov kho los ntawm lub cim xeeb ntawm cov node uas tau hloov pauv, tso lub ntsuas phoo, tsis siv cov nplooj ntawv hloov kho hauv cov caches ntawm lwm cov nodes, thiab muab cov nplooj ntawv xauv rau cov node uas thov nws.

"Dirty", uas yog, hloov pauv, nplooj ntawv tuaj yeem sau rau disk ob qho tib si los ntawm qhov tsis tu ncua thiab los ntawm PowerHA (castout).

Yog tias ib qho ntawm pureScale nodes ua tsis tiav, kev rov qab tsuas yog txwv rau cov kev lag luam uas tseem tsis tau ua tiav thaum lub sijhawm ua tsis tiav: cov nplooj ntawv hloov kho los ntawm cov node hauv kev ua tiav yog nyob rau hauv lub ntiaj teb cache ntawm PowerHA. Cov node restarts nyob rau hauv ib tug txo configuration ntawm ib tug ntawm cov servers nyob rau hauv lub pawg, rolls rov qab nyob rau hauv kev lag luam thiab tso locks.

PowerHA khiav ntawm ob lub servers thiab tus tswv node replicates nws lub xeev synchronously. Yog tias thawj PowerHA node ua tsis tiav, pawg txuas ntxiv ua haujlwm nrog rau qhov thaub qab.
Tau kawg, yog tias koj nkag mus rau cov ntaub ntawv teeb tsa los ntawm ib qho ntawm ib qho, tag nrho kev ua haujlwm ntawm pawg yuav siab dua. PureScale tseem tuaj yeem pom tias qee thaj chaw ntawm cov ntaub ntawv tab tom ua tiav los ntawm ib qho, thiab tom qab ntawd tag nrho cov xauv cuam tshuam nrog thaj chaw ntawd yuav raug ua tiav hauv zos los ntawm cov node yam tsis muaj kev sib txuas lus nrog PowerHA. Tab sis sai li sai tau raws li daim ntawv thov sim nkag mus rau cov ntaub ntawv no los ntawm lwm qhov ntawm, kev ua haujlwm hauv nruab nrab ntawm lub xauv yuav rov pib dua.

IBM cov kev ntsuam xyuas sab hauv ntawm kev ua haujlwm ntawm 90% nyeem thiab 10% sau, uas zoo ib yam li cov khoom tsim tawm hauv ntiaj teb tiag tiag, qhia yuav luag cov kab sib tw mus txog 128 ntawm. Kev kuaj mob, hmoov tsis, tsis qhia tawm.

HPE NonStop SQL

Lub Hewlett-Packard Enterprise portfolio kuj muaj nws tus kheej muaj ntau platform. Qhov no yog NonStop platform, tso tawm rau kev ua lag luam xyoo 1976 los ntawm Tandem Computers. Nyob rau hauv 1997, lub tuam txhab tau txais los ntawm Compaq, uas nyob rau hauv lem merged nrog Hewlett-Packard nyob rau hauv 2002.

NonStop yog siv los tsim cov ntawv thov tseem ceeb - piv txwv li, HLR lossis bank card ua. Lub platform yog xa mus rau hauv daim ntawv ntawm software thiab kho vajtse complex (kwv yees), uas suav nrog kev suav cov nodes, cov ntaub ntawv khaws cia thiab cov khoom siv sib txuas lus. ServerNet network (hauv cov tshuab niaj hnub no - Infiniband) ua haujlwm rau ob qho tib si rau kev sib pauv ntawm cov nodes thiab nkag mus rau cov ntaub ntawv khaws cia.

Thaum ntxov versions ntawm lub system siv proprietary processors uas tau synchronized nrog ib leeg: tag nrho cov kev khiav hauj lwm tau ua synchronously los ntawm ob peb processors, thiab sai li sai tau raws li ib tug ntawm cov processors ua yuam kev, nws twb muab tua, thiab lub thib ob txuas ntxiv mus ua hauj lwm. Tom qab ntawd, lub kaw lus tau hloov mus rau cov txheej txheem sib xyaw ua ke (thawj MIPS, tom qab ntawd Itanium thiab thaum kawg x86), thiab lwm yam txheej txheem pib siv rau synchronization:

  • lus: txhua txheej txheem txheej txheem muaj "duab ntxoov ntxoo" ntxaib, uas cov txheej txheem nquag xa cov lus hais txog nws cov xwm txheej; yog tias cov txheej txheem tseem ceeb ua tsis tiav, cov txheej txheem duab ntxoov ntxoo pib ua haujlwm txij li lub sijhawm txiav txim los ntawm cov lus kawg;
  • kev pov npav: lub kaw lus khaws cia muaj cov khoom siv tshwj xeeb uas lees txais ntau qhov kev nkag tau zoo ib yam thiab ua tiav lawv tsuas yog tias qhov kev nkag mus sib tw; Hloov chaw ntawm lub cev synchronization, processors ua haujlwm asynchronously, thiab cov txiaj ntsig ntawm lawv txoj haujlwm tsuas yog piv rau lub sijhawm I / O.

Txij li xyoo 1987, kev sib raug zoo DBMS tau khiav ntawm NonStop platform - thawj SQL / MP, thiab tom qab SQL / MX.

Tag nrho cov ntaub ntawv tau muab faib ua ntu, thiab txhua ntu yog lub luag haujlwm rau nws tus kheej Cov Ntaub Ntawv Nkag Nkag (DAM) txheej txheem. Nws muab cov ntaub ntawv kaw, caching, thiab xauv cov txheej txheem. Kev ua cov ntaub ntawv yog ua los ntawm Executor Server Processes khiav ntawm tib lub nodes raws li cov ntaub ntawv tswj hwm. SQL / MX teem sijhawm faib cov haujlwm ntawm cov neeg ua haujlwm thiab sib sau cov txiaj ntsig. Thaum tsim nyog los ua kom pom zoo hloov pauv, ob-theem kev cog lus raws tu qauv muab los ntawm TMF (Kev Tswj Xyuas Kev Lag Luam) lub tsev qiv ntawv yog siv.

Distributed DBMS rau Enterprise

NonStop SQL tuaj yeem ua qhov tseem ceeb rau cov txheej txheem kom ntev analytical queries tsis cuam tshuam nrog kev ua lag luam. Txawm li cas los xij, nws lub hom phiaj yog qhov tseeb ntawm kev ua haujlwm luv luv, thiab tsis yog kev tshuaj xyuas. Tus tsim tawm lav qhov muaj ntawm NonStop pawg ntawm qib tsib "nine", uas yog, lub sijhawm poob qis tsuas yog 5 feeb hauv ib xyoos.

SAB HANA

Thawj qhov kev tso tawm ruaj khov ntawm HANA DBMS (1.0) tau tshwm sim thaum Lub Kaum Ib Hlis 2010, thiab pob SAP ERP hloov mus rau HANA thaum lub Tsib Hlis 2013. Lub platform yog raws li kev yuav khoom: TREX Search Engine (nrhiav hauv columnar cia), P * TIME DBMS thiab MAX DB.

Lo lus "HANA" nws tus kheej yog ib lub ntsiab lus, Kev ua haujlwm siab analytical Appliance. DBMS no tau muab rau hauv daim ntawv code uas tuaj yeem khiav ntawm ib qho x86 servers, txawm li cas los xij, kev teeb tsa kev lag luam tau tso cai rau ntawm cov khoom siv tau ntawv pov thawj nkaus xwb. Cov kev daws teeb meem muaj los ntawm HP, Lenovo, Cisco, Dell, Fujitsu, Hitachi, NEC. Qee qhov kev teeb tsa Lenovo txawm tso cai rau kev ua haujlwm yam tsis muaj SAN - lub luag haujlwm ntawm cov txheej txheem khaws cia yog ua los ntawm GPFS pawg ntawm cov disks hauv zos.

Tsis zoo li cov platforms teev saum toj no, HANA yog ib lub cim xeeb DBMS, piv txwv li cov ntaub ntawv thawj cov duab khaws cia hauv RAM, thiab tsuas yog cov ntawv teev cia thiab cov duab thaij duab ib ntus raug sau rau disk kom rov qab tau thaum muaj kev puas tsuaj.

Distributed DBMS rau Enterprise

Txhua HANA pawg node yog lub luag haujlwm rau nws tus kheej ib feem ntawm cov ntaub ntawv, thiab daim ntawv qhia cov ntaub ntawv khaws cia rau hauv ib qho tshwj xeeb - Lub Npe Server, nyob rau ntawm tus neeg sib koom tes. Cov ntaub ntawv tsis yog duplicated ntawm nodes. Cov ntaub ntawv xauv kuj tseem khaws cia rau ntawm txhua qhov, tab sis lub kaw lus muaj lub ntsuas ntsuas thoob ntiaj teb.

Thaum tus neeg siv khoom HANA txuas mus rau ib pawg, nws rub tawm nws cov topology thiab tuaj yeem nkag mus rau txhua qhov ncaj qha, nyob ntawm seb nws xav tau cov ntaub ntawv dab tsi. Yog tias kev hloov pauv cuam tshuam rau cov ntaub ntawv ntawm ib qho ntawm ib qho, ces nws tuaj yeem ua tiav hauv zos los ntawm cov node, tab sis yog tias cov ntaub ntawv ntawm ob peb lub nodes hloov pauv, qhov pib ntawm node hu rau tus neeg koom tes ntawm node, uas qhib thiab tswj cov kev faib tawm, ua nws siv ib qho optimized ob-theem commit raws tu qauv.

Lub coordinator node yog duplicated, yog li yog hais tias tus coordinator ua tsis tau tejyam, lub backup node yuav siv sij hawm tam sim ntawd. Tab sis yog hais tias ib tug node nrog cov ntaub ntawv tsis ua hauj lwm, ces tib txoj kev mus saib tau nws cov ntaub ntawv yog restart lub node. Raws li txoj cai, HANA pawg tswj hwm cov neeg siv khoom seem kom rov pib qhov poob ntawm nws sai li sai tau.

Tau qhov twg los: www.hab.com

Ntxiv ib saib