Inani elikhulu lezinhlelo ze-Enterprise kanye nezinhlelo ze-virtualization zinezindlela zazo zokwakha izixazululo ezibekezelela amaphutha. Ngokukhethekile, i-Oracle RAC (I-Oracle Real Application Cluster) iyiqoqo lamaseva esizindalwazi se-Oracle amabili noma ngaphezulu asebenza ndawonye ukuze alinganisele umthwalo futhi anikeze ukubekezelelana kwamaphutha ezingeni leseva/lesicelo. Ukuze usebenze kule modi, udinga isitoreji esabiwe, ngokuvamile okuyisistimu yokugcina.
Njengoba sesike saxoxa kwenye yethu , isistimu yesitoreji ngokwayo, naphezu kokuba khona kwezingxenye eziphindwe kabili (kuhlanganise nezilawuli), kusenamaphuzu okuhluleka - ikakhulukazi ngesimo sesethi eyodwa yedatha. Ngakho-ke, ukuze kwakhiwe isixazululo se-Oracle esinezidingo ezengeziwe zokuthembeka, uhlelo lwe-“N amaseva - uhlelo olulodwa lwesitoreji” ludinga ukuba inkimbinkimbi.
Okokuqala, yiqiniso, kudingeka sinqume ukuthi yiziphi izingozi esizama ukuziqinisekisa ngazo. Kulesi sihloko, ngeke sicabangele ukuvikeleka ezinsongweni ezifana nokuthi “i-meteorite isifikile.” Ngakho ukwakha isixazululo sokutakula izinhlekelele esihlakazekile kuzohlala kuyisihloko kwesinye sezindatshana ezilandelayo. Lapha sizobheka lokho okubizwa ngokuthi yi-Cross-Rack isixazululo sokubuyisela inhlekelele, lapho ukuvikelwa kwakhiwa ezingeni lamakhabhinethi eseva. Amakhabhinethi ngokwawo angatholakala ekamelweni elilodwa noma ahlukene, kodwa ngokuvamile ngaphakathi kwesakhiwo esifanayo.
Lawa makhabhinethi kufanele aqukathe yonke isethi edingekayo yemishini kanye nesofthiwe ezovumela ukusebenza kolwazi lwe-Oracle kungakhathaliseki ukuthi "umakhelwane" unjani. Ngamanye amazwi, sisebenzisa isixazululo sokutakula inhlekelele ye-Cross-Rack, sisusa ubungozi bokwehluleka:
- Amaseva wohlelo lwe-Oracle
- Izinhlelo zokugcina
- Ukushintsha amasistimu
- Ukuhluleka okuphelele kwazo zonke izinto zokusebenza kukhabhinethi:
- Ukwenqaba amandla
- Ukuhluleka kwesistimu yokupholisa
- Izici zangaphandle (umuntu, imvelo, njll.)
Ukuphindwaphindwa kwamaseva e-Oracle kusho umgomo wokusebenza we-Oracle RAC futhi usetshenziswa ngohlelo lokusebenza. Ukuphindwaphindwa kwezindawo zokushintshwa nakho akuyona inkinga. Kodwa ngokuphindaphinda kwesistimu yokugcina, yonke into ayilula kangako.
Inketho elula ukuphindaphinda kwedatha kusuka ohlelweni lwesitoreji oluyinhloko kuya kweyokulondoloza. I-synchronous noma i-asynchronous, kuye ngamakhono esistimu yokugcina. Ngokuphindaphinda okungavumelaniyo, umbuzo uvela ngokushesha wokuqinisekisa ukuvumelana kwedatha maqondana ne-Oracle. Kodwa noma ngabe kukhona ukuhlanganiswa kwesofthiwe nesicelo, kunoma yikuphi, uma kwenzeka ukwehluleka ohlelweni oluyinhloko lokugcina, ukungenelela okwenziwa ngabaphathi kuyodingeka ukuze kushintshwe iqoqo kusitoreji sokusekelayo.
Inketho eyinkimbinkimbi kakhulu isofthiwe kanye/noma isitoreji se-hardware "i-virtualizers" ezoqeda izinkinga zokungaguquguquki nokungenelela okwenziwa ngesandla. Kodwa inkimbinkimbi yokuthunyelwa kanye nokuphatha okulandelayo, kanye nezindleko ezingafaneleki kakhulu zezixazululo ezinjalo, kwethusa abaningi.
Isixazululo se-AccelStor NeoSapphire™ All Flash array silungele izimo ezifana nokululama kwenhlekelele ye-Cross-Rack usebenzisa i-Shared-Nothing architecture. Le modeli iwuhlelo lokugcina olunamanodi amabili olusebenzisa ubuchwepheshe bobunikazi be-FlexiRemap® ukuze isebenze ngama-flash drive. Sibonga u I-NeoSapphire™ H710 iyakwazi ukuletha ukusebenza okungafika kokungu-600K IOPS@4K kokubhala okungahleliwe nokufundwa okungahleliwe okungu-1M+ IOPS@4K, okungafinyeleleki uma kusetshenziswa amasistimu esitoreji asekelwe ku-RAID.
Kodwa isici esiyinhloko se-NeoSapphire™ H710 ukukhishwa kwamanodi amabili ngendlela yamakesi ahlukene, ngalinye elinekhophi yalo yedatha. Ukuvumelanisa ama-node kwenziwa ngokusebenzisa i-interface ye-InfiniBand yangaphandle. Ngenxa yalesi sakhiwo, kungenzeka ukusabalalisa ama-node ezindaweni ezahlukene ebangeni elingafika ku-100m, ngaleyo ndlela unikeze isisombululo sokubuyisela inhlekelele ye-Cross-Rack. Womabili ama-node asebenza ngokuhambisana ngokuphelele. Kusukela ohlangothini lomsingathi, i-H710 ibukeka njengesistimu yokugcina yesilawuli esikabili. Ngakho-ke, asikho isidingo sokwenza noma iyiphi isofthiwe eyengeziwe noma izinketho zehadiwe noma izilungiselelo eziyinkimbinkimbi.
Uma siqhathanisa zonke izixazululo zokubuyisela inhlekelele ye-Cross-Rack ezichazwe ngenhla, khona-ke inketho evela ku-AccelStor igqama ngokuphawulekayo kwezinye:
I-AccelStor NeoSapphire™ Yabelana Ngezinto Zokwakha
Isofthiwe noma isistimu yokugcina ye-hardware “virtualizer”
Isixazululo esisekelwe ekuphindaphindeni
Ukutholakala
Ukuhluleka kweseva
Asikho Isikhathi Sokuphumula
Asikho Isikhathi Sokuphumula
Asikho Isikhathi Sokuphumula
Ukuhluleka kokushintsha
Asikho Isikhathi Sokuphumula
Asikho Isikhathi Sokuphumula
Asikho Isikhathi Sokuphumula
Ukuhluleka kwesistimu yesitoreji
Asikho Isikhathi Sokuphumula
Asikho Isikhathi Sokuphumula
Isikhathi sokuphumula
Ukwehluleka okuphelele kwekhabhinethi
Asikho Isikhathi Sokuphumula
Asikho Isikhathi Sokuphumula
Isikhathi sokuphumula
Izindleko nobunkimbinkimbi
Izindleko zesixazululo
Phansi*
Высокая
Высокая
Ubunkimbinkimbi bokuthunyelwa
Okuphansi
Высокая
Высокая
*I-AccelStor NeoSapphire™ iseyi-All Flash array, leyo ngokwencazelo engabizi "3 kopecks," ikakhulukazi njengoba inomthamo ophindwe kabili. Kodwa-ke, uma uqhathanisa izindleko zokugcina zesisombululo esisekelwe kuso nezifanayo ezivela kwabanye abathengisi, izindleko zingabhekwa njengeziphansi.
I-topology yokuxhuma amaseva ohlelo lokusebenza kanye nawo Wonke ama-Flash array node azobukeka kanje:
Lapho uhlela i-topology, kubuye kunconywe kakhulu ukuphinda amaswishi okuphatha kanye namaseva okuxhumana.
Lapha futhi ngokuqhubekayo sizokhuluma ngokuxhuma nge-Fiber Channel. Uma usebenzisa i-iSCSI, yonke into izofana, ilungiselwe izinhlobo zokushintshwa ezisetshenzisiwe kanye nezilungiselelo zamalungu afanayo ahluke kancane.
Umsebenzi wokulungiselela kuhlu
Izisetshenziswa nesoftware esetshenzisiwe
Iseva futhi Shintsha Imininingwane
Izingxenye
Incazelo
I-Oracle Database 11g amaseva
Amabili
Uhlelo lokusebenza lweseva
I-Oracle Linux
Inguqulo ye-Oracle database
11g (RAC)
Amaphrosesa ngeseva ngayinye
Ama-cores amabili angu-16 Intel® Xeon® CPU E5-2667 v2 @ 3.30GHz
Inkumbulo yomzimba ngeseva ngayinye
128GB
Inethiwekhi ye-FC
I-16Gb/s FC ene-multipathing
FC HBA
I-Emulex Lpe-16002B
Izimbobo ezinikelwe zomphakathi ze-1GbE zokuphatha iqoqo
I-adaptha ye-Intel Ethernet RJ45
16Gb/s FC switch
I-Brocade 6505
Izimbobo ezizinikezele ze-10GbE eziyimfihlo zokuvumelanisa idatha
I-Intel X520
I-AccelStor NeoSapphire™ Konke Ukucaciswa Kwe-Flash Array
Izingxenye
Incazelo
Uhlelo lokugcina
Imodeli yokutholakala ephezulu ye-NeoSapphire™: H710
Inguqulo yesithombe
4.0.1
Isamba senani lamadrayivu
48
Usayizi weDrayivu
1.92TB
Uhlobo lweDrayivu
I-SSD
I-FC target port
16x 16Gb izimbobo (8 inodi ngayinye)
Izimbobo zokuphatha
Ikhebula ye-ethernet engu-1GbE exhuma kubasingathi ngeswishi ye-ethernet
Imbobo yokushaya kwenhliziyo
Ikhebula ye-ethernet engu-1GbE exhuma phakathi kwamanodi esitoreji amabili
Imbobo yokuvumelanisa idatha
Intambo ye-InfiniBand engu-56Gb/s
Ngaphambi kokuthi usebenzise uhlelo, kufanele uluqalise. Ngokuzenzakalelayo, ikheli lokulawula lawo womabili ama-node liyafana (192.168.1.1). Udinga ukuxhuma kubo ngamunye ngamunye bese usetha amakheli okuphatha amasha (asevele ahlukile) futhi usethe ukuvumelanisa isikhathi, ngemva kwalokho amachweba Wokuphatha angaxhunywa kunethiwekhi eyodwa. Ngemva kwalokho, ama-node ahlanganiswa abe yipheya ye-HA ngokunikeza ama-subnets okuxhumana kwe-Interlink.
Ngemuva kokuthi ukuqalisa sekuqediwe, ungaphatha amalungu afanayo kunoma iyiphi indawo.
Okulandelayo, sakha amavolumu adingekayo futhi siwashicilele kumaseva ohlelo lokusebenza.
Kutuswa kakhulu ukuthi udale amavolumu amaningi e-Oracle ASM njengoba lokhu kuzokhuphula inani lezinhloso zamaseva, okuzogcina kuthuthukise ukusebenza okuphelele (okuningi emigqeni kwenye ).
Ukucushwa kokuhlola
Igama levolumu yesitoreji
Usayizi Wevolumu
Idatha01
200GB
Idatha02
200GB
Idatha03
200GB
Idatha04
200GB
Idatha05
200GB
Idatha06
200GB
Idatha07
200GB
Idatha08
200GB
Idatha09
200GB
Idatha10
200GB
I-Grid01
1GB
I-Grid02
1GB
I-Grid03
1GB
I-Grid04
1GB
I-Grid05
1GB
I-Grid06
1GB
Yenza futhi01
100GB
Yenza futhi02
100GB
Yenza futhi03
100GB
Yenza futhi04
100GB
Yenza futhi05
100GB
Yenza futhi06
100GB
Yenza futhi07
100GB
Yenza futhi08
100GB
Yenza futhi09
100GB
Yenza futhi10
100GB
Ezinye izincazelo mayelana nezindlela zokusebenza zamalungu afanayo nezinqubo ezenzeka ezimeni eziphuthumayo
Isethi yedatha yenodi ngayinye inepharamitha "yenombolo yenguqulo". Ngemuva kokuqaliswa kokuqala, kuyafana futhi kulingana no-1. Uma ngesizathu esithile inombolo yenguqulo ihlukile, khona-ke idatha ihlale ivumelaniswa kusukela kunguqulo endala kuya kwencane, ngemva kwalokho inombolo yenguqulo encane ihambisana, i.e. lokhu kusho ukuthi amakhophi ayefana. Izizathu zokuthi kungani izinguqulo zingahluka:
- Ukuqalisa kabusha okuhleliwe kwenye yamanodi
- Ingozi kwenye yezindawo ngenxa yokuvalwa okungazelelwe (ukunikezwa kwamandla, ukushisa ngokweqile, njll.).
- Kulahleke uxhumano lwe-InfiniBand ngokungakwazi ukuvumelanisa
- Ukuphahlazeka kwenye yamanodi ngenxa yokonakala kwedatha. Lapha uzodinga ukudala iqembu elisha le-HA futhi uqedele ukuvumelanisa kwesethi yedatha.
Kunoma ikuphi, inodi ehlala iku-inthanethi inyusa inombolo yenguqulo yayo ngeyodwa ukuze ivumelanise isethi yayo yedatha ngemva kokubuyiselwa kokuxhumana nokubhanqwa.
Uma ukuxhumeka ngesixhumanisi se-Ethernet kulahleka, i-Heartbeat ishintshela ku-InfiniBand okwesikhashana bese ibuya phakathi kwamasekhondi angu-10 lapho isibuyiselwe.
Isetha ababungazi
Ukuqinisekisa ukubekezelelwa kwamaphutha futhi uthuthukise ukusebenza, kufanele unike amandla ukusekelwa kwe-MPIO kwamalungu afanayo. Ukuze wenze lokhu, udinga ukwengeza imigqa efayeleni /etc/multipath.conf, bese uqala kabusha isevisi ye-multipath.
Umbhalo ofihliweamadivayisi {
idivayisi {
umthengisi "Astor"
path_grouping_policy "group_by_prio"
isikhethi_sendlela "ubude bomugqa 0"
path_checker "tur"
izici "0"
hardware_isibambi "0"
i-prio "const"
ukwehluleka ngokushesha
fast_io_fail_tmo 5
dev_loss_tmo 60
user_friendly_names yebo
detect_prio yebo
rr_min_io_rq 1
no_path_retry 0
}
}
Okulandelayo, ukuze i-ASM isebenze ne-MPIO nge-ASMLib, udinga ukushintsha ifayela /etc/sysconfig/oracleasm bese ugijima /etc/init.d/oracleasm scandisks
Umbhalo ofihliwe
# ORACLEASM_SCANORDER: Ukufanisa amaphethini ukuze uhlele ukuskena kwediski
ORACLEASM_SCANORDER="dm"
# ORACLEASM_SCANEXCLUDE: Ukufanisa amaphethini ukuze kukhishwe amadiski ekuskeneni
ORACLEASM_SCANEXCLUDE="sd"
Ukubhala
Uma ungafuni ukusebenzisa i-ASMLib, ungasebenzisa imithetho ye-UDEV, eyisisekelo se-ASMLib.
Kusukela ngenguqulo 12.1.0.2 ye-Oracle Database, inketho iyatholakala ukuze ifakwe njengengxenye yesofthiwe ye-ASMFD.
Kubalulekile ukuqinisekisa ukuthi amadiski adalelwe i-Oracle ASM ahambisana nosayizi webhulokhi uhlelo olusebenza ngawo ngokomzimba (4K). Uma kungenjalo, izinkinga zokusebenza zingase zenzeke. Ngakho-ke, kuyadingeka ukudala amavolumu ngemingcele efanele:
kuhlukanisiwe /dev/mapper/igama ledivayisi mklabel gpt mkpart primary 2048s 100% qondanisa-hlola okuphelele 1
Ukusatshalaliswa kwemininingo egciniwe kuwo wonke amavolumu adaliwe wokucushwa kokuhlolwa kwethu
Igama levolumu yesitoreji
Usayizi Wevolumu
Imephu yevolumu ye-LUNs
Imininingwane Yedivayisi Yevolumu ye-ASM
Usayizi Weyunithi Yokwabiwa
Idatha01
200GB
Mepha wonke amavolumu esitoreji ohlelweni lwesitoreji kuzo zonke izimbobo zedatha
I-redundancy: Normal
Igama:DGDATA
Inhloso:Amafayela edatha
4MB
Idatha02
200GB
Idatha03
200GB
Idatha04
200GB
Idatha05
200GB
Idatha06
200GB
Idatha07
200GB
Idatha08
200GB
Idatha09
200GB
Idatha10
200GB
I-Grid01
1GB
I-redundancy: Normal
Igama: DGGRID1
Inhloso:Igridi: I-CRS kanye Nokuvota
4MB
I-Grid02
1GB
I-Grid03
1GB
I-Grid04
1GB
I-redundancy: Normal
Igama: DGGRID2
Inhloso:Igridi: I-CRS kanye Nokuvota
4MB
I-Grid05
1GB
I-Grid06
1GB
Yenza futhi01
100GB
I-redundancy: Normal
Igama: DGREDO1
Inhloso: Phinda wenze ilogi yochungechunge 1
4MB
Yenza futhi02
100GB
Yenza futhi03
100GB
Yenza futhi04
100GB
Yenza futhi05
100GB
Yenza futhi06
100GB
I-redundancy: Normal
Igama: DGREDO2
Inhloso: Phinda wenze ilogi yochungechunge 2
4MB
Yenza futhi07
100GB
Yenza futhi08
100GB
Yenza futhi09
100GB
Yenza futhi10
100GB
Izilungiselelo Zesizindalwazi
- Usayizi wokuvimba = 8K
- Shintsha isikhala = 16GB
- Khubaza i-AMM (Automatic Memory Management)
- Khubaza Amakhasi Amakhulu Asobala
Ezinye izilungiselelo
# vi /etc/sysctl.conf
✓ fs.aio-max-nr = 1048576
✓ fs.file-max = 6815744
✓ kernel.shmmax 103079215104
✓ kernel.shmall 31457280
✓ kernel.shmmn 4096
✓ kernel.sem = 250 32000 100 128
✓ net.ipv4.ip_local_port_range = 9000 65500
✓ net.core.rmem_default = 262144
✓ net.core.rmem_max = 4194304
✓ net.core.wmem_default = 262144
✓ net.core.wmem_max = 1048586
✓vm.swappiness=10
✓ vm.min_free_kbytes=524288 # ungakusethi lokhu uma usebenzisa i-Linux x86
✓ vm.vfs_cache_pressure=200
✓ vm.nr_hugepages = 57000
# vi /etc/security/limits.conf
✓ igridi ethambile nproc 2047
✓ igridi eqinile nproc 16384
✓ igridi ethambile ye-nofile 1024
✓ i-grid hard nofile 65536
✓ isitaki esithambile segridi 10240
✓ isitaki esiqinile segridi 32768
✓ i-oracle soft nproc 2047
✓ i-oracle kanzima nproc 16384
✓ i-oracle soft nofile 1024
✓ i-oracle hard nofile 65536
✓ isitaki esithambile se-oracle 10240
✓ isitaki esiqinile se-oracle 32768
✓ i-memlock ethambile 120795954
✓ i-hard memlock 120795954
sqlplus "/ as sysdba"
shintsha isethi yesistimu=2000 scope=spfile;
shintsha isistimu isethi open_cursors=2000 scope=spfile;
shintsha isethi yesistimu session_cached_cursors=300 scope=spfile;
shintsha isethi yesistimu db_files=8192 scope=spfile;
Ukuhlolwa kokwehluleka
Ngezinjongo zokubonisa, i-HammerDB isetshenziselwe ukulingisa umthwalo we-OLTP. Ukucushwa kwe-HammerDB:
Inombolo Yezinqolobane
256
Inani Lokwenziwe ngomsebenzisi ngamunye
1000000000000
Abasebenzisi Ababonakalayo
256
Umphumela ube yi-2.1M TPM, okude kakhulu nomkhawulo wokusebenza wezinhlu , kodwa "iwuphahla" lokucushwa kwehadiwe yamanje yamaseva (ngokuyinhloko ngenxa yamaphrosesa) kanye nenombolo yawo. Inhloso yalokhu kuhlolwa kusewukukhombisa ukubekezelela iphutha kwesixazululo lilonke, hhayi ukufeza ukusebenza okuphezulu. Ngakho-ke, sizomane sakhe phezu kwalesi sibalo.
Hlola ukwehluleka kwelinye lamanodi
Ababungazi balahlekelwe ingxenye yezindlela eziya endaweni yokugcina izinto, beqhubeka nokusebenza ngalezo ezisele nge-node yesibili. Ukusebenza kwehle imizuzwana embalwa ngenxa yezindlela ezakhiwa kabusha, kwase kubuyiselwa kokujwayelekile. Akubanga khona ukuphazamiseka enkonzweni.
Ukuhlolwa kokuhluleka kweKhabhinethi ngazo zonke izinto zokusebenza
Kulokhu, ukusebenza kuphinde kwehla imizuzwana embalwa ngenxa yokuhlelwa kabusha kwezindlela, kwase kubuyiselwa kuhhafu yenani lokuqala. Umphumela ube uhhafu kusukela kowokuqala ngenxa yokukhishwa kweseva yohlelo lokusebenza olulodwa ekusebenzeni. Akubanga nasiphazamiso nakusevisi.
Uma kunesidingo sokuqalisa isixazululo esibekezelela iphutha se-Cross-Rack sokutakula izinhlekelele se-Oracle ngezindleko ezizwakalayo kanye nomzamo omncane wokuphakela/wokuphatha, i-Oracle RAC kanye nezakhiwo kusebenza ndawonye. kuzoba enye yezinketho ezinhle kakhulu. Esikhundleni se-Oracle RAC, kungaba nanoma iyiphi enye isofthiwe ehlinzeka ngokuhlanganisa, i-DBMS efanayo noma izinhlelo ze-virtualization, isibonelo. Umgomo wokwakha isixazululo uzohlala unjalo. Futhi okubalulekile kunguziro we-RTO ne-RPO.
Source: www.habr.com