{"id":54686,"date":"2020-01-01T00:00:00","date_gmt":"2019-12-31T21:00:00","guid":{"rendered":"https:\/\/prohoster.info\/blog\/blog_prohoster\/5-8-mln-iops-zachem-tak-mnogo"},"modified":"2020-02-18T14:02:44","modified_gmt":"2020-02-18T11:02:44","slug":"5-8-mln-iops-zachem-tak-mnogo","status":"publish","type":"post","link":"https:\/\/prohoster.info\/et\/blog\/administrirovanie\/5-8-mln-iops-zachem-tak-mnogo","title":{"rendered":"5.8 miljonit IOPS: miks nii palju?","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Tere Habr! Andmekogumid Big Data ja masin\u00f5ppe jaoks kasvavad eksponentsiaalselt ja tuleb j\u00f5uda nende t\u00f6\u00f6tlemiseni. Meie postitus on veel \u00fche innovaatilise tehnoloogia kohta k\u00f5rgetasemelistes arvutustes (HPC, High Performance Computing), mis oli v\u00e4lja pandud Kingstonis Supercomputing-2019<\/a><\/noindex>. See on Hi-End andmesalvestuss\u00fcsteemide (\u0421\u0425\u0414) kasutamine graafikaprotsessoritega (GPU) serverites ja GPUDirect Storage tehnoloogia. Otsese andmevahetuse t\u00f5ttu \u0421\u0425\u0414 ja GPU vahel, m\u00f6\u00f6da CPU-d, kiireneb andmete laadimine GPU-kiirenditesse mitmekordselt, mist\u00f5ttu Big Data rakendused t\u00f6\u00f6tavad maksimaalse j\u00f5udlusega, mille tagavad GPU-d. Omakorda huvitavad HPC-s\u00fcsteemide arendajaid saavutused \u0421\u0425\u0414 valdkonnas, mis tagavad \u00fclimalt k\u00f5rge sisendi\/v\u00e4ljundi kiirus \u2014 sellised, nagu toodab Kingston.<\/p>\n

\"5.8
\n<\/a><\/noindex><\/p>\n

GPU j\u00f5udlus \u00fcletab andmete laadimist <\/h2>\n

\nAlates CUDA loomist 2007. aastal \u2014 GPU-l p\u00f5hineva paralleelse arvutusarhitektuuri, mis on m\u00f5eldud \u00fcldotstarbeliste rakenduste arendamiseks, on GPU riistvaralised v\u00f5imalused tohutult kasvanud. T\u00e4na leiduvad GPU-d \u00fcha enam HPC-rakendustes, nagu suured andmed (Big Data), masin\u00f5pe (ML, machine learning) ja s\u00fcva\u00f5pe (DL, deep learning). <\/p>\n

Kuigi terminid on sarnased, on kaks viimast algoritmiliselt erinevad \u00fclesanded. ML \u00f5petab arvutit struktureeritud andmete p\u00f5hjal, samas kui DL toetub neuroniv\u00f5rgu tagasisidele. N\u00e4ide, mis aitab m\u00f5ista erinevusi, on \u00fcsna lihtne. Oletame, et arvuti peab eristama fotosid kassidest ja koertest, mis laaditakse sisse HCI-st. ML jaoks tuleks esitada pildikollektsioon paljude siltidega, kus iga\u00fcks m\u00e4\u00e4ratleb m\u00f5ne loomade omaduse. DL-i puhul on piisavalt laadida palju rohkem pilte, kuid ainult \u00fche sildi \u201esee on kass\u201c v\u00f5i \u201esee on koer\u201c all. DL sarnaneb sellele, kuidas v\u00e4ikesi lapsi \u00f5petatakse \u2014 neile lihtsalt n\u00e4idatakse koerte ja kasside pilte raamatutes ja elus (enamasti isegi, nagu ei seletataks detailset erinevust), ja lapse aju hakkab iseseisvalt tuvastama looma t\u00fc\u00fcpi p\u00e4rast teatud kriitilise arvu v\u00f5rdluspilte (hinnanguliselt r\u00e4\u00e4gime k\u00f5igest sajast-kahek\u00fcmnest n\u00e4itamisest kogu varases lapsep\u00f5lves). DL algoritmid pole veel nii t\u00e4iuslikud: et neuroniv\u00f5rk suudaks edukalt kujutiste m\u00e4\u00e4ramisel t\u00f6\u00f6tada, on vajalik edastada ja t\u00f6\u00f6delda GPU-s miljoneid pilte.<\/p>\n

Eelneva kokkuv\u00f5te: GPU-p\u00f5hised HPC rakendused Big Data, ML ja DL valdkonnas on v\u00f5imalikud, kuid probleemiks on, et andmekogud on nii suured, et andmete laadimise aeg salvestusseadmest GPU-sse hakkab v\u00e4hendama rakenduse \u00fcldist j\u00f5udlust. Teisis\u00f5nu, kiired graafikaprotsessorid j\u00e4\u00e4vad alakoormatud aeglaste sisend-\/v\u00e4ljadatumise t\u00f5ttu, mis tulevad teistelt alams\u00fcsteemidelt. GPU ja CPU\/HCI busside sisendi\/v\u00e4ljundi kiirus v\u00f5ib olla k\u00fcmneid kordi erinev.<\/p>\n

Kuidas t\u00f6\u00f6tab tehnoloogia GPUDirect Storage?<\/h2>\n

\nSisendi\/v\u00e4ljundi protsessi kontrollib CPU, samuti ka andmete laadimise protsess salvestusseadmest graafikaprotsessoritesse edasiseks t\u00f6\u00f6tlemiseks. Seet\u00f5ttu tekkis vajadus tehnoloogia j\u00e4rele, mis tagaks otsese ligip\u00e4\u00e4su GPU-de ja NVMe-diskide vahel kiireks suhtlemiseks. Esimese sellise tehnoloogia pakkus NVIDIA ja nimetas selle GPUDirect Storage-iks. P\u00f5him\u00f5tteliselt on see eelnevalt v\u00e4lja t\u00f6\u00f6tatud tehnoloogia GPUDirect RDMA (Remote Direct Memory Address) variatsioon.<\/p>\n

\"5.8
\nJensen Huang, NVIDIA peadirektor, esitleb GPUDirect Storage'it GPUDirect RDMA variandina n\u00e4itusel SC-19. Allikas: NVIDIA<\/i><\/p>\n

GPUDirect RDMA ja GPUDirect Storage vahe on seadmetes, mille vahel aadresseerimine toimub. GPUDirect RDMA tehnoloogia on m\u00f5eldud andmete edastamiseks otse v\u00f5rguliidese (NIC) sisendkaardilt GPU m\u00e4lu, samas kui GPUDirect Storage tagab otse\u00fchenduse andmete edastamiseks kohaliku v\u00f5i eemal asuva salvestusseadmest, nagu NVMe v\u00f5i NVMe over Fabric (NVMe-oF), ja GPU m\u00e4luni.<\/p>\n

M\u00f5lemad variandid, GPUDirect RDMA ja GPUDirect Storage, v\u00e4ldivad liigsete andmete liikumisi CPU m\u00e4lu puhver kaudu ja v\u00f5imaldavad otsep\u00e4\u00e4su m\u00e4lu mehhanismil (DMA) edastada andmeid otse v\u00f5rgu- v\u00f5i salvestuskaardilt GPU m\u00e4llu v\u00f5i tagasiviimise, ilma et keskpink oleks koormatud. GPUDirect Storage puhul ei oma salvestuskoha asukoht t\u00e4htsust: see v\u00f5ib olla NVME-disk GPU-ga seadmes, rack'is v\u00f5i v\u00f5rgu kaudu \u00fchendatud nagu NVMe-oF. <\/p>\n

\"5.8
\nGPUDirect Storage t\u00f6\u00f6p\u00f5him\u00f5te. Allikas: NVIDIA<\/i><\/p>\n

Hi-End NVMe SCSI-d on n\u00f5utud HPC-rakenduste turul<\/h2>\n

\nTuvastades, et GPUDirect Storage'i kasutuselev\u00f5tuga hakatakse suurte klientide t\u00e4helepanu p\u00f6\u00f6rama salvestuss\u00fcsteemide pakkumiseks, mille IO kiirus vastab GPU l\u00e4bilaskev\u00f5imele, n\u00e4itas Kingston messil SC-19 demo s\u00fcsteemi, mis koosnes NVMe-diskide p\u00f5hinevatest SCSI-dest ja GPU seadmetest, kus anal\u00fc\u00fcsiti tuhandeid satelliitpilte sekundis. Sellest SCSI-st k\u00fcmne DC1000M U.2 NVMe draiviga oleme juba kirjutanud superarvutite messil<\/a><\/noindex>. <\/p>\n

\"5.8
\nK\u00fcmne DC1000M U.2 NVMe draiviga SCSI t\u00e4iendab v\u00e4\u00e4rikalt graafikakiirendi serverit. Allikas: Kingston<\/i><\/p>\n

Selline SCSI on erinev 1U v\u00f5i suurema rack'i seadme kujul ning seda saab skaleerida vastavalt DC1000M U.2 NVMe ketaste arvule, kus iga\u00fche maht on 3.84-7.68 TB. DC1000M on esimene NVMe SSD mudel U.2 form-factor, mis kuulub Kingston'i andmekeskuste draivide seeriasse. Selle kestvuse reiting (DWPD, Drive writes per day) v\u00f5imaldab andmeid t\u00e4ielikult igal p\u00e4eval kord kirjutada garantii kehtivusaja jooksul. <\/p>\n

Testis fio v3.13 Ubuntu 18.04.3 LTS operatsioonis\u00fcsteemiga, Linux kernel 5.0.0-31-generic, n\u00e4itas n\u00e4idis SCSI lugemiskiirus (Sustained Read) 5.8 miljonit IOPS ja stabiilne l\u00e4bilaskev\u00f5ime (Sustained Bandwidth) 23.8 Gbit\/s.<\/p>\n

Ari\u00ebl Peres, \u00e4ri\u00fcrituste juht SSD Kingstonis, iseloomustas uusi andmesalvestusseadmmeid j\u00e4rgmiselt: \u201eOleme valmis varustama j\u00e4rgmise p\u00f5lvkonna servereid U.2 NVMe SSD lahendustega, et k\u00f5rvaldada paljud andmeedastusega seotud kitsaskohad, mis on traditsiooniliselt seotud salvestuss\u00fcsteemidega. NVMe SSD-draivide ja meie kvaliteetse Server Premier DRAM m\u00e4luga kombinatsioon muudab Kingstoni \u00fcheks k\u00f5ige t\u00e4ielikumaks lahenduste pakkujaks andmete t\u00f6\u00f6tlemise valdkonnas.\u201d<\/p>\n

\"5.8
\nGFIO v3.13 test n\u00e4itas 23,8 Gbit\/s l\u00e4bilaskev\u00f5imet DC1000M U.2 NVMe diskidel p\u00f5hineva demonstreerimise andmesalvestusseadmme jaoks. Allikas: Kingston<\/i><\/p>\n

Milline n\u00e4eb v\u00e4lja t\u00fc\u00fcpiline s\u00fcsteem HPC rakenduste jaoks, kus on rakendatud GPUDirect Storage tehnoloogiat v\u00f5i sarnaseid lahendusi? See on arhitektuur, millel on f\u00fc\u00fcsiline funktsionaalsete plokkide eraldatus riiulis: \u00fcks-kaks \u00fcksust m\u00e4lule, veel m\u00f5ned GPU ja CPU arvutusss\u00f5lmedele ning \u00fcks v\u00f5i mitu \u00fcksust andmesalvestusseadmmele.<\/p>\n

GPUDirect Storage ja teiste GPU tootjate sarnaste tehnoloogiate tutvustamisega kasvab Kingstoniseks n\u00f5udlus HPC-dele suunatud andmesalvestusseadmmete j\u00e4rele. M\u00e4rgiks on andmete lugemise kiirus andmesalvestusseadmmetest, mis on v\u00f5rreldav 40- v\u00f5i 100-Gbit\/s v\u00f5rgukaartide l\u00e4bilaskek\u00f5rgenderiga GPU arvutus\u00fcksuses. Seega, ultra-k\u00f5rge kiirus, sealhulgas v\u00e4lised NVMe kaudu Fabric, muutuvad eksootilisest mainstreamiks HPC rakendustes. Lisaks teadusele ja finantsarvutustele leiavad need rakendust paljudes muudes praktilistes valdkondades, n\u00e4iteks megapoli turvas\u00fcsteemides Safe City v\u00f5i transpordivalve keskustes, kus on vajalik tuvastamise ja tuvastamise kiirus, mis ulatub miljonite HD-piltide tasemele sekundis,\u201c m\u00e4rkis tippkategooria andmesalvestusseadmmete turuni\u0161i.<\/p>\n

Lisainfo Kingston'i toodete kohta leiate ametlikul kodulehe<\/a><\/noindex> ettev\u00f5ttega.
\n
Allikas: habr.com<\/a><\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"

\u041f\u0440\u0438\u0432\u0435\u0442 \u0425\u0430\u0431\u0440! \u041d\u0430\u0431\u043e\u0440\u044b \u0434\u0430\u043d\u043d\u044b\u0445 \u0434\u043b\u044f Big Data \u0438 \u043c\u0430\u0448\u0438\u043d\u043d\u043e\u0433\u043e \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f \u044d\u043a\u0441\u043f\u043e\u043d\u0435\u043d\u0446\u0438\u0430\u043b\u044c\u043d\u043e \u0440\u0430\u0441\u0442\u0443\u0442 \u0438 \u043d\u0430\u0434\u043e \u0443\u0441\u043f\u0435\u0432\u0430\u0442\u044c \u0438\u0445 \u043e\u0431\u0440\u0430\u0431\u0430\u0442\u044b\u0432\u0430\u0442\u044c. \u041d\u0430\u0448 \u043f\u043e\u0441\u0442 \u043e \u0435\u0449\u0435 \u043e\u0434\u043d\u043e\u0439 \u0438\u043d\u043d\u043e\u0432\u0430\u0446\u0438\u043e\u043d\u043d\u043e\u0439 \u0442\u0435\u0445\u043d\u043e\u043b\u043e\u0433\u0438\u0438 \u0432 \u043e\u0431\u043b\u0430\u0441\u0442\u0438 \u0432\u044b\u0441\u043e\u043a\u043e\u043f\u0440\u043e\u0438\u0437\u0432\u043e\u0434\u0438\u0442\u0435\u043b\u044c\u043d\u044b\u0445 \u0432\u044b\u0447\u0438\u0441\u043b\u0435\u043d\u0438\u0439 (HPC, High Performance Computing), \u043f\u043e\u043a\u0430\u0437\u0430\u043d\u043d\u043e\u0439 \u043d\u0430 \u0441\u0442\u0435\u043d\u0434\u0435 Kingston \u043d\u0430 Supercomputing-2019. \u042d\u0442\u043e \u043f\u0440\u0438\u043c\u0435\u043d\u0435\u043d\u0438\u0435 Hi-End \u0441\u0438\u0441\u0442\u0435\u043c \u0445\u0440\u0430\u043d\u0435\u043d\u0438\u044f \u0434\u0430\u043d\u043d\u044b\u0445 (\u0421\u0425\u0414) \u0432 \u0441\u0435\u0440\u0432\u0435\u0440\u0430\u0445 \u0441 \u0433\u0440\u0430\u0444\u0438\u0447\u0435\u0441\u043a\u0438\u043c\u0438 \u043f\u0440\u043e\u0446\u0435\u0441\u0441\u043e\u0440\u0430\u043c\u0438 (GPU) \u0438 \u0442\u0435\u0445\u043d\u043e\u043b\u043e\u0433\u0438\u0435\u0439 \u0448\u0438\u043d\u044b GPUDirect […]<\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[688],"tags":[],"class_list":["post-54686","post","type-post","status-publish","format-standard","hentry","category-administrirovanie"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\n","aioseo_head_json":{"title":"\ud83e\udd475,8 miljonit IOPS: miks nii palju? | ProHoster","description":"Tere, Habr! Suurte andmete ja masin\u00f5ppe andmekogud kasvavad eksponentsiaalselt ning neid tuleb t\u00f6\u00f6delda.","canonical_url":"https:\/\/prohoster.info\/et\/blog\/administrirovanie\/5-8-mln-iops-zachem-tak-mnogo","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"et_EE","og:site_name":"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b","og:type":"article","og:title":"\ud83e\udd475.8 \u043c\u043b\u043d IOPS: \u0437\u0430\u0447\u0435\u043c \u0442\u0430\u043a \u043c\u043d\u043e\u0433\u043e? | ProHoster","og:description":"\u041f\u0440\u0438\u0432\u0435\u0442 \u0425\u0430\u0431\u0440! \u041d\u0430\u0431\u043e\u0440\u044b \u0434\u0430\u043d\u043d\u044b\u0445 \u0434\u043b\u044f Big Data \u0438 \u043c\u0430\u0448\u0438\u043d\u043d\u043e\u0433\u043e \u043e\u0431\u0443\u0447\u0435\u043d\u0438\u044f \u044d\u043a\u0441\u043f\u043e\u043d\u0435\u043d\u0446\u0438\u0430\u043b\u044c\u043d\u043e \u0440\u0430\u0441\u0442\u0443\u0442 \u0438 \u043d\u0430\u0434\u043e \u0443\u0441\u043f\u0435\u0432\u0430\u0442\u044c \u0438\u0445 \u043e\u0431\u0440\u0430\u0431\u0430\u0442\u044b\u0432\u0430\u0442\u044c.","og:url":"https:\/\/prohoster.info\/et\/blog\/administrirovanie\/5-8-mln-iops-zachem-tak-mnogo","og:image":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:secure_url":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:width":350,"og:image:height":350,"article:published_time":"2019-12-31T21:00:00+00:00","article:modified_time":"2020-02-18T11:02:44+00:00","article:publisher":"https:\/\/www.facebook.com\/prohoster","article:author":"https:\/\/www.facebook.com\/prohoster"},"aioseo_meta_data":{"post_id":"54686","title":null,"description":null,"keywords":null,"keyphrases":null,"primary_term":null,"canonical_url":null,"og_title":null,"og_description":null,"og_object_type":"default","og_image_type":"default","og_image_url":null,"og_image_width":null,"og_image_height":null,"og_image_custom_url":null,"og_image_custom_fields":null,"og_video":null,"og_custom_url":null,"og_article_section":null,"og_article_tags":null,"twitter_use_og":false,"twitter_card":"default","twitter_image_type":"default","twitter_image_url":null,"twitter_image_custom_url":null,"twitter_image_custom_fields":null,"twitter_title":null,"twitter_description":null,"schema":{"blockGraphs":[],"customGraphs":[],"default":{"data":{"Article":[],"Course":[],"Dataset":[],"FAQPage":[],"Movie":[],"Person":[],"Product":[],"ProductReview":[],"Car":[],"Recipe":[],"Service":[],"SoftwareApplication":[],"WebPage":[]},"graphName":"","isEnabled":true},"graphs":[]},"schema_type":null,"schema_type_options":null,"pillar_content":false,"robots_default":true,"robots_noindex":false,"robots_noarchive":false,"robots_nosnippet":false,"robots_nofollow":false,"robots_noimageindex":false,"robots_noodp":false,"robots_notranslate":false,"robots_max_snippet":null,"robots_max_videopreview":null,"robots_max_imagepreview":"large","priority":null,"frequency":null,"local_seo":null,"seo_analyzer_scan_date":"2026-01-24 12:22:25","breadcrumb_settings":null,"limit_modified_date":false,"reviewed_by":null,"ai":null,"created":"2021-02-28 20:00:11","updated":"2026-01-24 12:22:25","focus_keyword":null,"additional_keywords":null,"truseo_locale":null},"gt_translate_keys":[{"key":"link","format":"url"}],"_links":{"self":[{"href":"https:\/\/prohoster.info\/et\/wp-json\/wp\/v2\/posts\/54686","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/et\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/et\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/et\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/et\/wp-json\/wp\/v2\/comments?post=54686"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/et\/wp-json\/wp\/v2\/posts\/54686\/revisions"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/et\/wp-json\/wp\/v2\/media?parent=54686"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/et\/wp-json\/wp\/v2\/categories?post=54686"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/et\/wp-json\/wp\/v2\/tags?post=54686"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}