Hoʻolālā ʻia ka ʻōnaehana Shenango e hoʻohana ʻia i nā kikowaena data.
/ kiʻi
Wahi a kekahi o nā mea hoʻolako, nā kikowaena data ʻo 20-40% wale nō o ka mana kamepiula i loaʻa. Ma nā haawe kiʻekiʻe kēia hōʻailona . ʻO kēia māhele o nā kumuwaiwai e alakaʻi i ka puka ʻana o nā mea i kapa ʻia ʻo "zombie servers". He mau mīkini kēia e noho wale i ka hapa nui o ka manawa, e hoʻopau i ka ikehu. I kēia lā 30% o nā kikowaena ma ka honua , e hoʻohana ana i $30 biliona waiwai o ka uila i kēlā me kēia makahiki.
Ua hoʻoholo ʻo MIT e hakakā i ka hoʻohana pono ʻole ʻana i nā kumuwaiwai computing.
Kime hana 'enekinia ʻōnaehana hoʻoili ukana i kapa ʻia ʻo Shenango. ʻO kāna kumu ka nānā ʻana i ke kūlana o ka pahu hana a hāʻawi hou i nā kaʻina hana paʻa (ʻaʻole hiki ke loaʻa ka manawa CPU) i nā mīkini manuahi.
Pehea e hana ai ʻo Shenango
ʻO Shenango kahi waihona Linux ma C me nā paʻa Rust a me C++. Hoʻopuka ʻia ka code papahana a me nā noi hoʻāʻo ma ma GitHub.
Hoʻokumu ʻia ka hoʻonā ma ka IOKernel algorithm, e holo ana ma luna o kahi kumu hoʻolaʻa o kahi ʻōnaehana multiprocessor. Mālama ia i nā noi CPU me ka hoʻohana ʻana i kahi framework , ka mea e hiki ai i nā noi ke kamaʻilio pololei me nā polokalamu pūnaewele.
Hoʻoholo ka IOKernel i nā kernels e hāʻawi i kahi hana kikoʻī. Hoʻoholo pū ka algorithm i ka nui o nā cores e pono ai. No kēlā me kēia kaʻina hana, ua hoʻoholo ʻia nā cores nui (hōʻoiaʻiʻo ʻia) a me nā mea hou aʻe (burstable) - hoʻokuʻu ʻia nā mea hope i ka wā e piʻi nui ai ka nui o nā noi i ka CPU.
Hoʻonohonoho ʻia ka laina noi IOKernel e like me . I kēlā me kēia ʻelima microseconds, nānā ka algorithm e ʻike inā ua pau nā hana a pau i hāʻawi ʻia i ke kumu. No ka hana ʻana i kēia, hoʻohālikelike ia i ka wahi o kēia manawa o ke poʻo o ka buffer me ke kūlana mua o kona huelo. Inā ʻike ʻia aia ka huelo i ka pila i ka manawa o ka nānā mua, ʻike ka ʻōnaehana i ka overload buffer a hoʻokaʻawale i kahi kumu hou no ke kaʻina hana.
I ka wā e puunaue ai i ka ukana, hāʻawi ʻia ka manaʻo nui i nā cores kahi i hana ʻia ai ke kaʻina hana like ma mua a waiho ʻia kekahi hapa i loko o ka cache, a i ʻole nā cores idle.
Lawe ʻo Shenango i ke ala . ʻO nā cores i hoʻokaʻawale ʻia e holo i hoʻokahi noi e nānā i ka nui o nā hana i loaʻa i kēlā me kēia. Inā hoʻopau kekahi kumu i kāna papa inoa o nā hana ma mua o nā mea ʻē aʻe, a laila "hoʻomaha" ia i kahi hapa o ka ukana mai kona mau hoalauna.
Nā pōmaikaʻi a me nā meaʻinoʻole
Nā ʻenehana mai MIT, hiki iā Shenango ke hana i ʻelima miliona mau noi i kēlā me kēia kekona a mālama i ka manawa pane awelika o 37 microseconds. Ua ʻōlelo ka poʻe loea i kekahi mau manawa hiki i ka ʻenehana ke hoʻonui i ka nui o ka hoʻohana ʻana o nā mea hana i nā kikowaena data i 100%. ʻO ka hopena, hiki i nā mea kikowaena data ke mālama i ke kūʻai ʻana a me ka mālama ʻana i nā kikowaena.
Hiki ke hoʻonā a me nā loea mai nā kulanui ʻē aʻe. Wahi a kekahi kumu aʻoaʻo mai kahi kula ʻo Korea, e kōkua ka ʻōnaehana MIT i ka hōʻemi ʻana i nā lohi i nā lawelawe pūnaewele. No ka laʻana, pono ia ma nā hale kūʻai pūnaewele. Ma nā lā kūʻai, aia ka lua o ka lohi i ka hoʻouka ʻana i ka ʻaoʻao i ka emi ʻana o ka helu o nā ʻike pūnaewele ma 11%. E kōkua ka hāʻawi ʻana i ka hoʻouka ʻana i nā mea kūʻai aku.
He mau hemahema ka ʻenehana - ʻaʻole ia e kākoʻo i ka multiprocessor - nā ʻōnaehana i hoʻopili ʻia nā chips i nā modula hoʻomanaʻo like ʻole a ʻaʻole "kamaʻilio" kekahi i kekahi. I kēia hihia, hiki iā IOKernel ke hoʻoponopono i ka hana o kahi hui kaʻawale o nā kaʻina hana, ʻaʻole naʻe nā chip server āpau.
/ kiʻi
Nā ʻenehana like
ʻO nā ʻōnaehana hoʻoili ukana ʻē aʻe me Arachne. Heluhelu ʻo ia i ka nui o nā cores e pono ai kahi noi i ka wā e hoʻomaka ai, a e puʻunaue i nā kaʻina hana e like me kēia hōʻailona. Wahi a nā mea kākau, ʻo ka latency kiʻekiʻe loa o kahi noi ma Arachne ma kahi o 10 tausani microseconds.
Hoʻokomo ʻia ka ʻenehana ma ke ʻano he waihona C++ no Linux, a loaʻa kāna code kumu ma .
ʻO kekahi mea hana kaulike ʻo ZygOS. E like me Shenango, hoʻohana ka ʻenehana i ke ʻano ʻaihue hana e hāʻawi hou i nā kaʻina hana. Wahi a nā mea kākau o ZygOS, ʻo ka latency noi maʻamau i ka wā e hoʻohana ai i ka mea hana ma kahi o 150 microseconds, a ʻo ka nui loa ma kahi o 450 microseconds. ʻO ke code papahana kekahi .
haʻina
Ke hoʻomau nei ka hoʻonui ʻana o nā kikowaena data hou. ʻIke ʻia ka piʻi ʻana o ke ʻano ma ka mākeke o nā kikowaena data hyperscale: i kēia manawa i ka honua. 430 mau kikowaena data hyperscale, akā i nā makahiki e hiki mai ana e piʻi aʻe ko lākou helu e 30%. No kēia kumu, e koi nui ʻia nā ʻenehana hoʻohālikelike ukana. Loaʻa nā ʻōnaehana e like me Shenango hui nui, a e ulu wale ana ka nui o ia mau mea paahana i keia mua aku.
Nā memo mai ka blog mua e pili ana i ka hui ʻo IaaS:
Source: www.habr.com