{"id":31148,"date":"2019-10-31T21:39:44","date_gmt":"2019-10-31T18:39:44","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novyj-balansirovshhik-nagruzki-na-cpu-ot-mit\/"},"modified":"2019-10-31T21:39:44","modified_gmt":"2019-10-31T18:39:44","slug":"novyj-balansirovshhik-nagruzki-na-cpu-ot-mit","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sq\/blog\/administrirovanie\/novyj-balansirovshhik-nagruzki-na-cpu-ot-mit","title":{"rendered":"Balancuesi i ri i ngarkes\u00ebs n\u00eb CPU nga MIT","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
Sistemi Shenango planifikohet t\u00eb p\u00ebrdoret n\u00eb qendrat e t\u00eb dh\u00ebnave.<\/h5>\n

\n\"Balancuesi<\/a><\/noindex>
\n\/ \u0444\u043e\u0442\u043e Marco Verch<\/a><\/noindex> CC BY<\/a><\/noindex><\/i><\/p>\n

Sipas nj\u00eb nga ofruesit, qendrat e t\u00eb dh\u00ebnave n\u00eb kompanin\u00eb Hecla Mining, e cila nxjerr argjend dhe metale t\u00eb tjera t\u00eb \u00e7muara. Kompjuteri \u00ebsht\u00eb instaluar n\u00eb nj\u00eb stacion n\u00eb Alaska, ku p\u00ebrdoret p\u00ebr t\u00eb vler\u00ebsuar parametrat e mineralit dhe p\u00ebr t\u00eb printuar etiketat p\u00ebr mostra. Me rastin e k\u00ebsaj, nj\u00eb printer po aq i vjet\u00ebr po p\u00ebrgjigjet p\u00ebr etiketa. \u00cbsht\u00eb interesante, se shtat\u00eb vjet m\u00eb par\u00eb, nj\u00ebri nga inxhinier\u00ebt e Hecla Mining n\u00eb nj\u00eb postim n\u00eb Reddit vuri n\u00eb dukje se \"ai nuk ka nevoj\u00eb t\u00eb luaj\u00eb serin\u00eb<\/a><\/noindex> kan\u00eb vet\u00ebm 20\u201340% t\u00eb kapaciteteve t\u00eb disponuesh\u00ebm t\u00eb p\u00ebrpunimit. N\u00eb ngarkesa t\u00eb larta, ky tregues mund t\u00eb arrij\u00eb deri n\u00eb 60%<\/a><\/noindex>. Nj\u00eb shp\u00ebrndarje e till\u00eb e burimeve \u00e7on n\u00eb shfaqjen e ashtuquajturave 'server\u00eb zombi'. K\u00ebto jan\u00eb makina q\u00eb kalojn\u00eb shumic\u00ebn e koh\u00ebs pa pun\u00eb, duke shpenzuar energji elektrike pa nevoj\u00eb. Sot, 30% e server\u00ebve n\u00eb bot\u00eb jan\u00eb t\u00eb papun\u00eb<\/a><\/noindex>, duke konsumuar elektricitet n\u00eb vler\u00ebn prej $30 miliard n\u00eb vit.<\/p>\n

P\u00ebr t\u00eb luftuar shpenzimin e panevojsh\u00ebm t\u00eb burimeve p\u00ebrpunuese, MIT ka vendosur<\/a><\/noindex><\/p>\n

Ekipi i inxhinier\u00ebve zhvilluan<\/a><\/noindex> nj\u00eb sistem balancimi ngarkese p\u00ebr procesor\u00ebt t\u00eb quajtur Shenango. Q\u00ebllimi i tij \u00ebsht\u00eb t\u00eb monitoroj\u00eb gjendjen e buffer-it t\u00eb detyrave dhe t\u00eb ribalancoj\u00eb proceset 'e bllokuara' (t\u00eb cilat nuk mund t\u00eb marrin koh\u00eb procesori) n\u00eb makina t\u00eb lira.<\/p>\n

Si funksionon Shenango<\/h2>\n

\nShenango \u00ebsht\u00eb nj\u00eb bibliotek\u00eb Linux n\u00eb gjuh\u00ebn C me lidhje t\u00eb Rust dhe C++. Kodi i projektit dhe aplikacionet testuese jan\u00eb publikuar n\u00eb repozitorit<\/a><\/noindex> n\u00eb GitHub.<\/p>\n

Baza e zgjidhjes \u00ebsht\u00eb algoritmi IOKernel, i cili ekzekutohet n\u00eb nj\u00eb b\u00ebrtham\u00eb t\u00eb p\u00ebrkushtuar t\u00eb nj\u00eb sistemi shum\u00ebprocesor. Ai menaxhon k\u00ebrkesat p\u00ebr CPU me ndihm\u00ebn e nj\u00eb framework-u DPDK<\/a><\/noindex>, i cili u jep mund\u00ebsin\u00eb aplikacioneve t\u00eb nd\u00ebrveprojn\u00eb drejtp\u00ebrdrejt me pajisjet rrjetit.<\/p>\n

IOKernel vendos se cilave b\u00ebrtha t'u d\u00ebrgoj\u00eb nj\u00eb detyr\u00eb specifike. Algoritmi gjithashtu vendos se sa b\u00ebrtha do t\u00eb nevojiten. P\u00ebr \u00e7do proces p\u00ebrcaktohen b\u00ebrthamat e garantuara dhe ato t\u00eb dukshme \u2014 t\u00eb dytat aktivizohen n\u00eb rast t\u00eb nj\u00eb rritjeje t\u00eb papritur t\u00eb numrit t\u00eb k\u00ebrkesave p\u00ebr CPU.<\/p>\n

Radhitja e k\u00ebrkesave t\u00eb IOKernel \u00ebsht\u00eb organizuar n\u00eb form\u00ebn e nj\u00eb buffer-i t\u00eb p\u00ebrhersh\u00ebm<\/a><\/noindex>. \u00c7do pes\u00eb mikrosekonda, algoritmi kontrollon n\u00ebse t\u00eb gjitha detyrat e caktuara p\u00ebr b\u00ebrtham\u00ebn jan\u00eb p\u00ebrfunduar. P\u00ebr k\u00ebt\u00eb, ai krahasohet me pozicionin aktual 't\u00eb kok\u00ebs' s\u00eb buffer-it me pozicionin e m\u00ebparsh\u00ebm t\u00eb 'bishtit' t\u00eb tij. N\u00ebse rezulton se bishti ka qen\u00eb n\u00eb radh\u00eb n\u00eb momentin e kontrollit t\u00eb m\u00ebparsh\u00ebm, sistemi sh\u00ebnon nj\u00eb mbingarkes\u00eb n\u00eb buffer dhe alokon nj\u00eb b\u00ebrtham\u00eb t\u00eb p\u00ebrkushtuar p\u00ebr procesin.<\/p>\n

Kur shp\u00ebrndahet ngarkesa, p\u00ebrpar\u00ebsi i jepet b\u00ebrthamave mbi t\u00eb cilat ky proces \u00ebsht\u00eb ekzekutuar m\u00eb par\u00eb dhe pjes\u00ebrisht \u00ebsht\u00eb mbetur n\u00eb cache, ose \u00e7do b\u00ebrtham\u00eb q\u00eb \u00ebsht\u00eb e papunuar.<\/p>\n

\"Balancuesi
\n
\nShenango gjithashtu p\u00ebrdor qasje vjedhje pune<\/a><\/noindex>. B\u00ebrthat q\u00eb jan\u00eb caktuar p\u00ebr t\u00eb punuar nj\u00eb aplikacion, monitorojn\u00eb numrin e detyrave midis tyre. N\u00ebse nj\u00eb b\u00ebrtham\u00eb p\u00ebrfundon list\u00ebn e detyrave m\u00eb her\u00ebt se t\u00eb tjerat, ajo 'mer' nj\u00eb pjes\u00eb t\u00eb ngarkes\u00ebs nga fqinj\u00ebt.<\/p>\n

Avantazhet dhe disavantazhet<\/h2>\n

\nSipas autoreve, sistemi do t\u00eb zvog\u00ebloj\u00eb shpenzimet p\u00ebr energjin\u00eb me dyfish.<\/a><\/noindex> ingenier\u00ebve nga MIT, Shenango \u00ebsht\u00eb n\u00eb gjendje t\u00eb procesoj\u00eb pes\u00eb milion k\u00ebrkesa n\u00eb sekond\u00eb dhe t\u00eb ruaj\u00eb nj\u00eb koh\u00eb t\u00eb mesme reagimi prej 37 mikrosekondash. Specialist\u00ebt thon\u00eb se n\u00eb disa raste teknologjia \u00ebsht\u00eb e aft\u00eb t\u00eb rris\u00eb shkall\u00ebn e p\u00ebrdorimit t\u00eb procesor\u00ebve n\u00eb qendrat e t\u00eb dh\u00ebnave deri n\u00eb 100%. Si pasoj\u00eb, operator\u00ebt e qendrave t\u00eb t\u00eb dh\u00ebnave do t\u00eb mund t\u00eb kursejn\u00eb n\u00eb blerjen dhe mir\u00ebmbajtjen e server\u00ebve.<\/p>\n

Potenciali i zgjidhjes theksojn\u00eb<\/a><\/noindex> dhe specialist\u00eb nga universitete t\u00eb tjera. Sipas nj\u00eb profesori nga nj\u00eb institut koreano, sistemi nga MIT do t\u00eb ndihmoj\u00eb n\u00eb reduktimin e vonesave n\u00eb funksionimin e sh\u00ebrbimeve t\u00eb uebit. P\u00ebr shembull, ai do t\u00eb jet\u00eb i dobish\u00ebm n\u00eb operimin e dyqaneve online. N\u00eb dit\u00ebt e shitjeve, madje edhe nj\u00eb vones\u00eb e vog\u00ebl e ngarkes\u00ebs s\u00eb faqes \u00e7on<\/a><\/noindex> \u00e7on n\u00eb nj\u00eb ulje prej 11% t\u00eb numrit t\u00eb shikimeve t\u00eb faqes.<\/p>\n

Shp\u00ebrndarja e menj\u00ebhershme e ngarkesave do t\u00eb ndihmoj\u00eb p\u00ebr t\u00eb sh\u00ebrbyer m\u00eb shum\u00eb klient\u00eb. NUMA<\/a><\/noindex>Me teknologjin\u00eb ende ka disavantazhe \u2014 ajo nuk mb\u00ebshtet shum\u00ebprocesor\u00ebt<\/p>\n

\"Balancuesi
\n\/ \u0444\u043e\u0442\u043e Tim Reckmann<\/a><\/noindex> CC BY<\/a><\/noindex><\/i><\/p>\n

-sistemat, n\u00eb t\u00eb cilat \u00e7ipat jan\u00eb t\u00eb lidhur me module t\u00eb ndryshme t\u00eb memories dhe nuk 'diskutojn\u00eb' nd\u00ebrmjet tyre. N\u00eb k\u00ebt\u00eb rast, IOKernel mund t\u00eb rregulloj\u00eb funksionimin e nj\u00eb grupi t\u00eb ve\u00e7ant\u00eb t\u00eb procesor\u00ebve, por jo t\u00eb gjith\u00eb \u00e7ipat e serverit.<\/h2>\n

\nTeknologji t\u00eb ngjashme<\/p>\n

Mes sistemeve t\u00eb tjera t\u00eb balancimit t\u00eb ngarkes\u00ebs s\u00eb procesor\u00ebve, mund t\u00eb p\u00ebrmendet Arachne. Ajo llogarit se sa b\u00ebrtha do t\u00eb nevojiten p\u00ebr nj\u00eb aplikacion n\u00eb momentin e nisjes, dhe shp\u00ebrndan proceset p\u00ebrkat\u00ebsisht. Sipas autor\u00ebve, vonesa maksimale e pun\u00ebs s\u00eb aplikacionit n\u00eb Arachne \u00ebsht\u00eb rreth 10 mij\u00eb mikrosekonda. GitHub<\/a><\/noindex>.<\/p>\n

Teknologjia \u00ebsht\u00eb implementuar n\u00eb form\u00ebn e nj\u00eb biblioteke C++ p\u00ebr Linux, dhe kodi i saj burimor \u00ebsht\u00eb n\u00eb Nj\u00eb tjet\u00ebr mjet balancues \u2014 ZygOS. Ashtu si Shenango, teknologjia p\u00ebrdor p\u00ebr shp\u00ebrndarjen e proceseve metod\u00ebn e vjedhjes s\u00eb pun\u00ebs. Sipas autor\u00ebve t\u00eb ZygOS, vonesa mesatare n\u00eb pun\u00ebn e aplikacioneve gjat\u00eb p\u00ebrdorimit t\u00eb instrumentit \u00ebsht\u00eb rreth 150 mikrosekonda, dhe maksimumi \u2014 rreth 450 mikrosekonda. Kodi i projektit gjithashtu<\/a><\/noindex>.<\/p>\n

P\u00ebrfundimet<\/h2>\n

\n\u00ebsht\u00eb n\u00eb qasje t\u00eb hapur Qendrat e t\u00eb dh\u00ebnave moderne vazhdojn\u00eb t\u00eb zgjerohen, sidomos tendenca e rritjes \u00ebsht\u00eb evidente n\u00eb tregun e qendrave t\u00eb t\u00eb dh\u00ebnave hyperscale: tani n\u00eb bot\u00eb<\/noindex> ekzistojn\u00eb 430 qendra t\u00eb dh\u00ebnash hyper-skal\u00eb, por brenda disa viteve numri i tyre mund t\u00eb rritet me 30%. P\u00ebr k\u00ebt\u00eb arsye, teknologjit\u00eb e balancimit t\u00eb ngarkes\u00ebs s\u00eb procesor\u00ebve do t\u00eb jen\u00eb shum\u00eb t\u00eb k\u00ebrkuara. Sistemet, t\u00eb ngjashme me Shenango, tashm\u00eb jan\u00eb<\/a><\/noindex> korporatat e m\u00ebdha, dhe n\u00eb t\u00eb ardhmen numri i k\u00ebtyre instrumenteve do t\u00eb rritet vet\u00ebm.<\/p>\n

\nPostime nga Blogu i Par\u00eb p\u00ebr IaaS t\u00eb korporatave:<\/p>\n