An rubuta wannan labarin don taimaka maka zaɓar mafita mai kyau don kanka da fahimtar bambance-bambance tsakanin SDS kamar Gluster, Ceph da Vstorage (Virtuozzo).
Rubutun yana amfani da hanyoyin haɗi zuwa labarai tare da ƙarin cikakkun bayanai na wasu matsalolin, don haka kwatancin za su kasance a takaice kamar yadda zai yiwu, ta amfani da mahimman bayanai ba tare da ɓata lokaci ba da bayanan gabatarwa waɗanda za ku iya, idan kuna so, samun kansu akan Intanet.
A gaskiya ma, ba shakka, batutuwan da aka taso suna buƙatar sautunan rubutu, amma a cikin zamani na zamani mutane da yawa ba sa son karatu da yawa))), don haka za ku iya karantawa da sauri kuma kuyi zabi, kuma idan wani abu ya kasance. ba bayyananne ba, bi hanyoyin haɗin yanar gizo ko kalmomin da ba a sani ba google))), kuma wannan labarin yana kama da madaidaicin abin rufewa don waɗannan batutuwa masu zurfi, yana nuna cikawa - mahimman mahimman abubuwan kowane yanke shawara.
Gundura
Bari mu fara da Gluster, wanda masana'antun dandamali na hyperconverged ke amfani da su tare da SDS bisa tushen buɗe ido don mahalli mai kama-da-wane kuma ana iya samun su akan gidan yanar gizon RedHat a cikin sashin ajiya, inda zaku iya zaɓar daga zaɓuɓɓukan SDS guda biyu: Gluster ko Ceph.
Gluster ya ƙunshi tarin fassarori - sabis ɗin da ke aiwatar da duk ayyukan rarraba fayiloli, da sauransu. Brick sabis ne da ke ba da faifai ɗaya, Volume juzu'i ne (pool) wanda ke haɗa waɗannan tubalin. Na gaba yana zuwa sabis don rarraba fayiloli zuwa ƙungiyoyi ta amfani da aikin DHT (rabawar zanta). Ba za mu haɗa da sabis na Sharding a cikin bayanin ba tunda hanyoyin haɗin da ke ƙasa za su bayyana matsalolin da ke tattare da shi.
Lokacin rubutawa, ana adana duka fayil ɗin cikin bulo kuma ana rubuta kwafinsa lokaci guda zuwa tubali akan sabar ta biyu. Bayan haka, za a rubuta fayil na biyu zuwa rukuni na biyu na tubali biyu (ko fiye) akan sabobin daban-daban.
Idan fayilolin kusan girman ɗaya ne kuma ƙarar ta ƙunshi rukuni ɗaya kawai, to komai yana da kyau, amma a ƙarƙashin wasu sharuɗɗan matsaloli masu zuwa zasu tashi daga kwatancen:
- Ana amfani da sarari a ƙungiyoyi ba daidai ba, ya dogara da girman fayilolin kuma idan babu isasshen sarari a cikin rukuni don rubuta fayil, za ku sami kuskure, ba za a rubuta fayil ɗin ba kuma ba za a sake rarraba shi zuwa wani rukuni ba. ;
- lokacin rubuta fayil ɗaya, IO yana zuwa rukuni ɗaya kawai, sauran ba su da aiki;
- ba za ku iya samun IO na gaba ɗaya ƙarar lokacin rubuta fayil ɗaya ba;
- da kuma general ra'ayi ya dubi kasa m saboda da rashin rarraba bayanai a cikin tubalan, inda shi ne sauki don daidaita da warware matsalar uniform rarraba, kuma ba kamar yadda yanzu da dukan fayil shiga cikin wani block.
Daga bayanin hukuma mun kuma zo ga fahimtar cewa gluster yana aiki azaman ajiyar fayil a saman kayan aikin RAID na yau da kullun. An sami yunƙurin haɓakawa don yanke fayilolin (Sharding) cikin tubalan, amma duk wannan ƙari ne wanda ke haifar da asarar aiki akan tsarin gine-ginen da aka rigaya ya kasance, tare da amfani da irin waɗannan abubuwan da aka rarraba cikin 'yanci tare da iyakokin aiki kamar Fuse. Babu sabis na metadata, wanda ke iyakance aiki da ikon haƙuri na kuskure lokacin da ake rarraba fayiloli zuwa tubalan. Ana iya lura da mafi kyawun alamun aiki tare da tsarin "Rarraba Maimaitawa" kuma adadin nodes ya kamata ya zama aƙalla 6 don tsara ingantaccen kwafi na 3 tare da rarraba kaya mafi kyau.
Waɗannan binciken kuma suna da alaƙa da bayanin ƙwarewar mai amfani kuma idan aka kwatanta da , kuma akwai kuma bayanin gwanintar da ke haifar da fahimtar wannan tsari mafi inganci kuma mafi aminci
Hoton yana nuna nauyin rarraba lokacin rubuta fayiloli guda biyu, inda aka rarraba kwafin fayil ɗin farko a cikin sabobin uku na farko, waɗanda aka haɗa su zuwa rukunin ƙarar 0, kuma ana sanya kwafi uku na fayil na biyu akan rukuni na biyu juzu'i1 na uku. sabobin. Kowane uwar garken yana da diski ɗaya.
Ƙaddamarwa gaba ɗaya ita ce, za ku iya amfani da Gluster, amma tare da fahimtar cewa za a sami iyakoki a cikin aiki da haƙurin kuskure waɗanda ke haifar da matsaloli a ƙarƙashin wasu sharuɗɗan mafita mai rikitarwa, inda ake buƙatar albarkatun don ɗimbin ƙididdiga na mahallin kama-da-wane.
Hakanan akwai wasu alamun aikin Gluster waɗanda za'a iya cimma su ƙarƙashin wasu sharuɗɗa, iyakance ga
ceph
Yanzu bari mu kalli Ceph daga bayanan gine-ginen da na iya Akwai kuma kwatanta tsakanin , inda za ku iya gane nan da nan cewa yana da kyau a tura Ceph akan sabobin daban, tunda ayyukansa suna buƙatar duk kayan aikin da ke ƙarƙashin kaya.
gine mafi rikitarwa fiye da Gluster kuma akwai ayyuka kamar sabis na metadata, amma gabaɗayan abubuwan abubuwan da aka gyara suna da wahala sosai kuma ba su da sassauƙa sosai don amfani da shi a cikin hanyar haɓakawa. Ana adana bayanan a cikin tubalan, wanda ya fi dacewa, amma a cikin tsarin duk ayyuka (bangarori), akwai asara da latency a ƙarƙashin wasu kaya da yanayin gaggawa, misali mai zuwa.
Daga bayanin gine-ginen, zuciya shine CRUSH, godiya ga wanda aka zaɓi wurin adana bayanai. Na gaba ya zo PG - wannan shine mafi wahalar abstraction (ƙungiyar ma'ana) don fahimta. Ana buƙatar PGs don sa CRUSH ya fi tasiri. Babban manufar PG shine tara abubuwa don rage yawan amfani da albarkatu, haɓaka aiki da haɓakawa. Magance abubuwa kai tsaye, daban-daban, ba tare da haɗa su cikin PG ba zai yi tsada sosai. OSD sabis ne na kowane faifai.
Tari na iya samun tafkunan bayanai ɗaya ko da yawa don dalilai daban-daban kuma tare da saituna daban-daban. An raba wuraren tafkuna zuwa ƙungiyoyin jeri. Ƙungiyoyin sanyawa suna adana abubuwan da abokan ciniki ke samu. Wannan shine inda matakin ma'ana ya ƙare, kuma matakin jiki ya fara, saboda kowane rukunin jeri an sanya shi babban faifai guda ɗaya da faifan kwafi da yawa (nawa daidai ya dogara da yanayin kwafin tafkin). A wasu kalmomi, a matakin ma'ana an adana abu a cikin takamaiman rukunin wuri, kuma a matakin jiki - akan faifai da aka sanya masa. A wannan yanayin, faifai na iya zama a zahiri a kan nodes daban-daban ko ma a cikin cibiyoyin bayanai daban-daban.
A cikin wannan makirci, ƙungiyoyin sanyawa suna kama da matakin da ya dace don sassaucin dukkanin bayani, amma a lokaci guda, a matsayin ƙarin hanyar haɗi a cikin wannan sarkar, wanda ba da gangan ba yana nuna asarar yawan aiki. Misali, lokacin rubuta bayanai, tsarin yana buƙatar raba su zuwa waɗannan ƙungiyoyi sannan a matakin zahiri cikin babban diski da diski don kwafi. Wato aikin Hash yana aiki ne yayin bincike da saka abu, amma akwai sakamako mai illa - yana da tsada sosai kuma yana hana sake gina zaren (lokacin ƙara ko cire diski). Wata matsalar zanta ita ce ƙusoshi a sarari na bayanan da ba za a iya canza su ba. Wato, idan ko ta yaya faifan yana cikin ƙarar nauyi, to tsarin ba ya da damar kada ya rubuta masa (ta hanyar zaɓar wani faifan), aikin hash yana wajabta adana bayanan bisa ga ka'ida, komai muni. faifan shine, don haka Ceph yana cin ƙwaƙwalwar ajiya mai yawa lokacin sake gina PG idan akwai warkar da kai ko haɓaka ajiya. Ƙarshen ita ce Ceph yana aiki da kyau (ko da yake a hankali), amma kawai lokacin da babu ƙima, yanayin gaggawa, ko sabuntawa.
Akwai, ba shakka, zaɓuɓɓuka don haɓaka aiki ta hanyar caching da raba cache, amma wannan yana buƙatar kayan aiki mai kyau kuma har yanzu za a sami asarar. Amma gabaɗaya, Ceph ya fi jaraba fiye da Gluster don yawan aiki. Har ila yau, lokacin amfani da waɗannan samfurori, wajibi ne a yi la'akari da wani muhimmin mahimmanci - wannan shine babban matakin ƙwarewa, ƙwarewa da ƙwarewa tare da babban mahimmanci akan Linux, tun da yake yana da mahimmanci don ƙaddamarwa, daidaitawa da kula da komai daidai. wanda ke ƙara ƙarin nauyi da nauyi akan mai gudanarwa.
Vstorage
Gine-ginen ya dubi mafi ban sha'awa , wanda za'a iya amfani dashi tare da hypervisor a kan nodes guda ɗaya, akan guda , amma yana da matukar muhimmanci a daidaita komai daidai don cimma kyakkyawan aiki. Wato, ƙaddamar da irin wannan samfurin daga akwatin akan kowane tsari ba tare da la'akari da shawarwarin da suka dace da tsarin gine-gine ba zai zama mai sauƙi, amma ba mai amfani ba.
Menene zai iya kasancewa tare don ajiya kusa da sabis na kvm-qemu hypervisor, kuma waɗannan ƴan sabis ne kawai inda aka sami ingantaccen matsayi na abubuwan haɗin gwiwa: sabis na abokin ciniki wanda aka saka ta FUSE (gyara, ba buɗaɗɗen tushe), sabis na metadata na MDS (Sabis na metadata), shingen bayanan sabis na Chunk, wanda a matakin jiki yayi daidai da faifai ɗaya kuma shi ke nan. Dangane da saurin gudu, ba shakka, yana da kyau a yi amfani da makirci mai jurewa kuskure tare da kwafi guda biyu, amma idan kuna amfani da caching da rajistan ayyukan akan faifan SSD, to za a iya rufe kurakuran-kuskure (shafe coding ko raid6) da kyau. tsarin matasan ko ma mafi kyau akan duk walƙiya. Akwai wasu hasara tare da EC (shafe coding): lokacin canza toshe bayanai guda ɗaya, dole ne a sake ƙididdige adadin adadin. Don keɓance asarar da ke tattare da wannan aikin, Ceph ya rubuta wa EC ba da daɗewa ba kuma matsalolin aiki na iya faruwa yayin takamaiman buƙatun, lokacin da, alal misali, duk tubalan suna buƙatar karantawa, kuma a cikin yanayin Virtuozzo Storage, ana aiwatar da rubuce-rubucen canza tubalan. ta amfani da tsarin "tsarin tsarin fayil", wanda ke rage ƙimar lissafin ƙididdiga. Don kimanta kusan zaɓuɓɓukan tare da haɓaka aiki tare da kuma ba tare da EC ba, akwai - alkalumman na iya zama kusan dangane da daidaiton ƙima na masana'antun kayan aiki, amma sakamakon ƙididdiga yana da taimako mai kyau wajen tsara tsarin.
Zane mai sauƙi na abubuwan ajiya ba yana nufin waɗannan abubuwan ba su sha ba amma idan kun lissafta duk farashin a gaba, zaku iya ƙidaya akan haɗin gwiwa kusa da hypervisor.
Akwai wani tsari don kwatanta amfani da kayan masarufi ta hanyar Ceph da sabis na ajiya na Virtuozzo.
Idan a baya yana yiwuwa a kwatanta Gluster da Ceph ta amfani da tsofaffin labaran, ta yin amfani da mahimman layi daga gare su, to tare da Virtuozzo ya fi wuya. Babu labarai da yawa akan wannan samfur kuma bayanai za'a iya tattarawa daga takaddun da ke kunne ko a cikin Rashanci idan muka yi la'akari da Vstorage a matsayin ajiya da aka yi amfani da shi a wasu hanyoyin magance rikice-rikice a cikin kamfanoni kamar da Acronis.
Zan yi ƙoƙarin taimakawa tare da bayanin wannan gine-gine, don haka za a sami ɗan ƙaramin rubutu, amma yana ɗaukar lokaci mai yawa don fahimtar takardun da kanku, kuma takaddun da ke akwai kawai za a iya amfani da su azaman tunani ta hanyar sake fasalin tebur. na abubuwan ciki ko bincike ta keyword.
Bari mu yi la'akari da tsarin rikodi a cikin ƙirar kayan masarufi tare da abubuwan da aka bayyana a sama: rikodi ya fara zuwa kullin da abokin ciniki ya fara shi (sabis ɗin Dutsen FUSE), amma sashin mahimmin Sabis na Metadata (MDS) zai ba shakka. kai tsaye abokin ciniki kai tsaye zuwa sabis ɗin da ake buƙata (sabis ɗin ajiya na CS tubalan), wato, MDS baya shiga cikin tsarin rikodi, amma kawai yana jagorantar sabis zuwa guntun da ake buƙata. Gabaɗaya, zamu iya ba da misalin yin rikodi tare da zuba ruwa a cikin ganga. Kowane ganga toshe bayanai 256MB ne.
Wato faifan guda ɗaya shine takamaiman adadin irin wannan ganga, wato ƙarar diski ɗin da aka raba zuwa 256MB. Ana rarraba kowane kwafin zuwa kumburi ɗaya, na biyu kusan a layi daya da wani kumburi, da sauransu ... Idan muna da kwafi guda uku kuma akwai diski na SSD don cache (don karantawa da rubuta rajistan ayyukan), to, tabbatar da rubutun zai faru bayan rubutawa. log ɗin zuwa SSD, kuma sake saitin layi ɗaya daga SSD zai ci gaba akan HDD, kamar a baya. A cikin yanayin kwafi uku, za a yi rikodin bayan tabbatarwa daga SSD na kumburi na uku. Yana iya zama alama cewa jimlar saurin rubuta SSDs uku za a iya raba uku kuma za mu sami saurin rubuta kwafi ɗaya, amma ana rubuta kwafin a layi daya kuma saurin Latency na cibiyar sadarwa yawanci ya fi na SSD, kuma a gaskiya aikin rubuta zai dogara ne akan hanyar sadarwa. Dangane da wannan, don ganin IOPS na ainihi, kuna buƙatar ɗaukar nauyin duka Vstorage daidai , wato, gwada ainihin kaya, kuma ba ƙwaƙwalwar ajiya da cache ba, inda ya zama dole don la'akari da girman toshe bayanai daidai, adadin zaren, da dai sauransu.
Littafin rikodin da aka ambata a sama akan SSD yana aiki ta yadda da zarar bayanai sun shiga ciki, sabis ɗin zai karanta shi nan da nan kuma a rubuta shi zuwa HDD. Akwai sabis na metadata da yawa (MDS) a kowane tari kuma adadin su yana ƙayyadadden ƙima, wanda ke aiki bisa ga Paxos algorithm. Daga ra'ayi na abokin ciniki, FUSE Dutsen batu shine babban fayil ɗin ajiya na cluster wanda ke gani a lokaci guda ga duk nodes a cikin cluster, kowane kumburi yana da madaidaicin abokin ciniki bisa ga wannan ka'ida, don haka wannan ajiyar yana samuwa ga kowane kumburi.
Don aiwatar da kowane ɗayan hanyoyin da aka bayyana a sama, yana da matukar mahimmanci, a matakin tsarawa da turawa, don daidaita hanyar sadarwar daidai, inda za a sami daidaitawa saboda tarawa da zaɓin bandwidth tashar cibiyar sadarwa daidai. A cikin tarawa, yana da mahimmanci a zaɓi yanayin hashing daidai da girman firam. Har ila yau, akwai bambanci mai ƙarfi daga SDS da aka kwatanta a sama, wannan fuse tare da fasahar hanya mai sauri a cikin Virtuozzo Storage. Wanne, ban da fis ɗin da aka sabunta, ba kamar sauran hanyoyin buɗe tushen ba, yana haɓaka IOPS sosai kuma yana ba ku damar iyakance ta hanyar sikeli ko a tsaye. Gabaɗaya, idan aka kwatanta da gine-ginen da aka bayyana a sama, wannan yana kama da ƙarfi, amma don irin wannan jin daɗi, ba shakka, kuna buƙatar siyan lasisi, sabanin Ceph da Gluster.
Don taƙaitawa, zamu iya haskaka saman ukun: Virtuozzo Storage yana ɗaukar matsayi na farko dangane da aiki da amincin gine-gine, Ceph yana matsayi na biyu, kuma Gluster yana matsayi na uku.
Ma'auni wanda aka zaɓi Virtuozzo Storage: shi ne mafi kyawun saiti na kayan aikin gine-gine, wanda aka sabunta don wannan tsarin Fuse tare da hanya mai sauri, tsari mai sassauƙa na daidaitawar kayan aiki, ƙarancin amfani da albarkatu da ikon rabawa tare da ƙididdigewa (ƙididdigewa / virtualization), wato, ya dace gaba daya don maganin hyperconverged , wanda ya kasance wani ɓangare na. Wuri na biyu shine Ceph saboda yana da kyakkyawan tsarin gine-gine idan aka kwatanta da Gluster, saboda aikinsa a cikin tubalan, da kuma mafi sassauƙan yanayi da ikon yin aiki a cikin manyan gungu.
Akwai shirye-shiryen rubuta kwatancen tsakanin vSAN, Space Direct Storage, Vstorage da Nutanix Storage, gwada Vstorage akan kayan HPE da Huawei, da kuma al'amuran don haɗawa da Vstorage tare da tsarin ajiyar kayan aiki na waje, don haka idan kuna son labarin, zai kasance. yana da kyau don samun ra'ayi daga gare ku, wanda zai iya ƙara ƙarfafawa don sababbin labaran, la'akari da sharhi da buri.
source: www.habr.com