{"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\/ro\/blog\/administrirovanie\/5-8-mln-iops-zachem-tak-mnogo","title":{"rendered":"5.8 milioane IOPS: de ce at\u00e2t de mult?","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Salut Habr! Seturile de date pentru Big Data \u0219i \u00eenv\u0103\u021barea automat\u0103 cresc exponential \u0219i trebuie s\u0103 ne adapt\u0103m pentru a le procesa. Postarea noastr\u0103 este despre o alt\u0103 tehnologie inovatoare \u00een domeniul calculului de \u00eenalt\u0103 performan\u021b\u0103 (HPC, High Performance Computing), prezentat\u0103 pe standul Kingston la Supercomputing-2019<\/a><\/noindex>. Aceasta este utilizarea sistemelor avansate de stocare (SAS) \u00een servere cu procesoare grafice (GPU) \u0219i tehnologia GPUDirect Storage. Datorit\u0103 schimbului de date direct \u00eentre SAS \u0219i GPU, ocolind CPU-ul, se accelereaz\u0103 semnificativ \u00eenc\u0103rcarea datelor \u00een acceleratoarele GPU, permi\u021b\u00e2nd aplica\u021biilor Big Data s\u0103 func\u021bioneze la capacitate maxim\u0103, oferit\u0103 de GPU-uri. \u00cen acela\u0219i timp, pe dezvoltatorii de sisteme HPC \u00eei intereseaz\u0103 progresele \u00een tehnologia SAS cu cele mai \u00eenalte viteze de intrare\/ie\u0219ire \u2014 cum ar fi cele produse de Kingston.<\/p>\n

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

Performan\u021ba GPU dep\u0103\u0219e\u0219te \u00eenc\u0103rcarea datelor <\/h2>\n

\nDe c\u00e2nd \u00een 2007 a fost creat CUDA \u2014 arhitectura hardware-software pentru calculul paralel bazat pe GPU pentru dezvoltarea aplica\u021biilor generice, capacit\u0103\u021bile hardware ale GPU-urilor au crescut incredibil. Ast\u0103zi, GPU-urile sunt tot mai utilizate \u00een aplica\u021bii HPC, cum ar fi Big Data, \u00eenv\u0103\u021barea automat\u0103 (ML, machine learning) \u0219i \u00eenv\u0103\u021barea profund\u0103 (DL, deep learning). <\/p>\n

Este de men\u021bionat c\u0103, \u00een ciuda similitudinii termenilor, ultimele dou\u0103 sunt probleme algoritmic diferite. ML \u00eenva\u021b\u0103 computerul pe baza datelor structurate, \u00een timp ce DL se bazeaz\u0103 pe r\u0103spunsul unei re\u021bele neuronale. Un exemplu care ajut\u0103 la \u00een\u021belegerea diferen\u021belor este destul de simplu. S\u0103 presupunem c\u0103 computerul trebuie s\u0103 disting\u0103 fotografiile c\u00e2inilor de cele ale pisicilor, care sunt \u00eenc\u0103rcate de pe un sistem de stocare. Pentru ML, este necesar s\u0103 se furnizeze un set de imagini cu multe etichete, fiecare determin\u00e2nd o anumit\u0103 caracteristic\u0103 a animalului. Pentru DL, este suficient s\u0103 \u00eenc\u0103rca\u021bi un num\u0103r mult mai mare de imagini, dar cu un singur etichet\u0103: \u201eaceasta este o pisic\u0103\u201d sau \u201eacesta este un c\u00e2ine\u201d. DL este foarte asem\u0103n\u0103tor cu modul \u00een care sunt \u00eenv\u0103\u021ba\u021bi copiii mici \u2014 ei sunt pur \u0219i simplu expu\u0219i la imagini ale c\u00e2inilor \u0219i pisicilor \u00een c\u0103r\u021bi \u0219i \u00een via\u021ba real\u0103 (de cele mai multe ori, f\u0103r\u0103 a explica detaliile diferen\u021bei), iar creierul copilului \u00eencepe singur s\u0103 disting\u0103 tipul animalului dup\u0103 un anumit num\u0103r critic de imagini pentru compara\u021bie (se estimeaz\u0103 c\u0103 este vorba despre o sut\u0103-dou\u0103 de expuneri pe tot parcursul copil\u0103riei timpurii). Algoritmii DL nu sunt \u00eenc\u0103 at\u00e2t de perfec\u021bi: pentru ca re\u021beaua neuronal\u0103 s\u0103 func\u021bioneze cu aceea\u0219i succes \u00een recunoa\u0219terea imaginilor, este necesar s\u0103 se furnizeze \u0219i s\u0103 se proceseze milioane de imagini \u00een GPU.<\/p>\n

Concluzia introducerii: pe baza GPU-urilor se pot construi aplica\u021bii HPC \u00een domeniul Big Data, ML \u0219i DL, dar exist\u0103 o problem\u0103 - seturile de date sunt at\u00e2t de mari \u00eenc\u00e2t timpul necesar pentru a \u00eenc\u0103rca datele din sistemul de stocare \u00een GPU \u00eencepe s\u0103 scad\u0103 performan\u021ba general\u0103 a aplica\u021biei. Cu alte cuvinte, procesoarele grafice rapide r\u0103m\u00e2n neutilizate din cauza vitezei lente de intrare-ie\u0219ire a datelor venind din alte subsisteme. Diferen\u021ba de vitez\u0103 de intrare-ie\u0219ire \u00eentre GPU \u0219i magistrala CPU\/Sistem de Stocare poate fi semnificativ\u0103.<\/p>\n

Cum func\u021bioneaz\u0103 tehnologia GPUDirect Storage?<\/h2>\n

\nProcesul de intrare-ie\u0219ire este controlat de CPU, la fel ca \u0219i procesul de \u00eenc\u0103rcare a datelor din stocare \u00een procesoarele grafice pentru procesare ulterioar\u0103. De aici apare cerin\u021ba pentru o tehnologie care s\u0103 ofere acces direct \u00eentre GPU \u0219i unit\u0103\u021bile NVMe pentru interac\u021biuni rapide \u00eentre ele. Prima astfel de tehnologie a fost propus\u0103 de NVIDIA \u0219i a fost numit\u0103 GPUDirect Storage. Practic, aceasta este o variant\u0103 a tehnologiei anterioare dezvoltate de ei, GPUDirect RDMA (Remote Direct Memory Address).<\/p>\n

\"5.8
\nJensen Huang, CEO-ul NVIDIA, prezint\u0103 GPUDirect Storage ca o variant\u0103 a GPUDirect RDMA la expozi\u021bia SC-19. Surs\u0103: NVIDIA<\/i><\/p>\n

Diferen\u021ba dintre GPUDirect RDMA \u0219i GPUDirect Storage const\u0103 \u00een dispozitivele \u00eentre care se face adresarea. Tehnologia GPUDirect RDMA este dedicat\u0103 mut\u0103rii datelor direct \u00eentre placa de re\u021bea (NIC) \u0219i memoria GPU, \u00een timp ce GPUDirect Storage ofer\u0103 un traseu direct pentru transmiterea datelor \u00eentre un depozit local sau la distan\u021b\u0103, cum ar fi NVMe sau NVMe over Fabric (NVMe-oF), \u0219i memoria GPU.<\/p>\n

Ambele op\u021biuni, GPUDirect RDMA \u0219i GPUDirect Storage, evit\u0103 mi\u0219c\u0103rile inutile ale datelor printr-un buffer \u00een memoria CPU \u0219i permit mecanismului de acces direct la memorie (DMA) s\u0103 transfere datele de la placa de re\u021bea sau depozit direct \u00een memoria GPU sau din aceasta \u2014 \u0219i toate acestea f\u0103r\u0103 a solicita procesorul central. \u00cen cazul GPUDirect Storage, loca\u021bia depozitului nu are importan\u021b\u0103: aceasta poate fi un disc NVME situat \u00een unitatea cu GPU, \u00een dulap sau conectat prin re\u021bea ca NVMe-oF. <\/p>\n

\"5.8
\nSchema de func\u021bionare a GPUDirect Storage. Sursa: NVIDIA<\/i><\/p>\n

Solu\u021biile de stocare Hi-End pe NVMe sunt foarte c\u0103utate pe pia\u021ba aplica\u021biilor HPC.<\/h2>\n

\n\u00cen\u021beleg\u00e2nd c\u0103, odat\u0103 cu introducerea GPUDirect Storage, interesul clien\u021bilor mari se va concentra pe ofertele de sisteme de stocare cu l\u0103\u021bimi de band\u0103 corespunz\u0103toare capacit\u0103\u021bii GPU, la t\u00e2rgul SC-19, Kingston a prezentat un demo al unui sistem format din solu\u021bii de stocare bazate pe discuri NVMe \u0219i unit\u0103\u021bi cu GPU, \u00een cadrul c\u0103ruia au fost analizate mii de imagini satelitare pe secund\u0103. Despre aceast\u0103 solu\u021bie de stocare bazat\u0103 pe 10 unit\u0103\u021bi DC1000M U.2 NVMe am mai scris. \u00een reportajul de la t\u00e2rgul supercomputerelor.<\/a><\/noindex>. <\/p>\n

\"5.8
\nSolu\u021bia de stocare bazat\u0103 pe 10 unit\u0103\u021bi DC1000M U.2 NVMe completeaz\u0103 \u00een mod adecvat un server cu acceleratoare grafice. Sursa: Kingston.<\/i><\/p>\n

Aceast\u0103 solu\u021bie de stocare este realizat\u0103 sub form\u0103 de unitate de rack de 1U sau mai mare \u0219i poate fi scalat\u0103 \u00een func\u021bie de num\u0103rul de discuri DC1000M U.2 NVMe, fiecare av\u00e2nd o capacitate de 3.84-7.68 TB. DC1000M este primul model NVMe SSD \u00een format U.2 din gama de solu\u021bii de stocare Kingston pentru centre de date. Acesta are o evaluare a durabilit\u0103\u021bii (DWPD, Drive writes per day), care permite rescrierea datelor pe \u00eentreaga capacitate o dat\u0103 pe zi pe parcursul duratei de via\u021b\u0103 garantate a dispozitivului. <\/p>\n

\u00cen testul fio v3.13 pe sistemul de operare Ubuntu 18.04.3 LTS, kernel Linux 5.0.0-31-generic, prototipul solu\u021biei de stocare a ar\u0103tat o vitez\u0103 de citire (Sustained Read) de 5.8 milioane IOPS cu o l\u0103\u021bime de band\u0103 sus\u021binut\u0103 (Sustained Bandwidth) de 23.8 Gbps.<\/p>\n

Ariel P\u00e9rez, manager de afaceri SSD la Kingston, a caracterizat noile solu\u021bii de stocare: \u201eSuntem preg\u0103ti\u021bi s\u0103 furniz\u0103m urm\u0103toarea genera\u021bie de servere cu solu\u021bii SSD U.2 NVMe, pentru a elimina multe dintre blocajele de transfer de date, care au fost \u00een mod tradi\u021bional asociate cu sistemele de stocare. Combina\u021bia dintre unit\u0103\u021bile NVMe SSD \u0219i memoria noastr\u0103 de calitate superioar\u0103 Server Premier DRAM face ca Kingston s\u0103 fie unul dintre cei mai completi furnizori din industrie de solu\u021bii integrate pentru procesarea datelor.\u201d<\/p>\n

\"5.8
\nTestul gfio v3.13 a ar\u0103tat o l\u0103\u021bime de band\u0103 de 23.8 Gbps pentru solu\u021bia de stocare demonstrativ\u0103 cu discuri DC1000M U.2 NVMe. Sursa: Kingston<\/i><\/p>\n

Cum va ar\u0103ta un sistem tipic pentru aplica\u021bii HPC, \u00een care este implementat\u0103 tehnologia GPUDirect Storage sau o tehnologie similar\u0103? Aceasta este o arhitectur\u0103 cu separarea fizic\u0103 a blocurilor func\u021bionale \u00een cadrul unit\u0103\u021bii: una sau dou\u0103 unit\u0103\u021bi pentru memorie, c\u00e2teva pentru noduri de calcul GPU \u0219i CPU \u0219i una sau mai multe unit\u0103\u021bi pentru solu\u021bia de stocare.<\/p>\n

Odat\u0103 cu anun\u021bul GPUDirect Storage \u0219i cu o posibil\u0103 apari\u021bie a tehnologiilor similare de la al\u021bi furnizori de GPU, pentru Kingston se extinde cererea de solu\u021bii de stocare concepute pentru utilizarea \u00een calculul de \u00eenalt\u0103 performan\u021b\u0103. Un indicator va fi viteza de citire a datelor din solu\u021biile de stocare, comparabil\u0103 cu l\u0103\u021bimea de band\u0103 a pl\u0103cilor de re\u021bea de 40 sau 100 Gbps la intrarea \u00een unitatea de calcul cu GPU. Astfel, solu\u021biile de stocare ultra-rapide, inclusiv NVMe externe prin Fabric, vor deveni mainstream pentru aplica\u021biile HPC. Pe l\u00e2ng\u0103 \u0219tiin\u021b\u0103 \u0219i calcule financiare, acestea vor g\u0103si aplica\u021bia \u00een multe alte domenii practice, cum ar fi sistemele de securitate la nivel de metropol\u0103 Safe City sau centrele de observa\u021bie \u00een transport, unde se cere viteza de recunoa\u0219tere \u0219i identificare la nivelul milioanelor de imagini HD pe secund\u0103\u201d, a definit ni\u0219a de pia\u021b\u0103 a celor mai bune solu\u021bii de stocare.<\/p>\n

Informa\u021bii suplimentare despre produsele Kingston pot fi g\u0103site pe site-ul oficial<\/a><\/noindex> companie.
\n
Sursa: 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 milioane IOPS: de ce at\u00e2ta nevoie?","description":"Salut, Habr! Seturile de date pentru Big Data \u0219i \u00eenv\u0103\u021barea automat\u0103 cresc exponen\u021bial, iar trebuie s\u0103 ne adapt\u0103m pentru a le procesa.","canonical_url":"https:\/\/prohoster.info\/ro\/blog\/administrirovanie\/5-8-mln-iops-zachem-tak-mnogo","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"ro_RO","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\/ro\/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\/ro\/wp-json\/wp\/v2\/posts\/54686","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/comments?post=54686"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/posts\/54686\/revisions"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/media?parent=54686"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/categories?post=54686"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/ro\/wp-json\/wp\/v2\/tags?post=54686"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}