Assalamu alaikum. Muna ci gaba da gabatar muku da tsarin ajiyar bayanai na Aerodisk VOSTOK, dangane da na'ura mai sarrafa Elbrus 8C ta Rasha.
A cikin wannan labarin mu (kamar yadda aka yi alkawari) za mu yi nazari dalla-dalla ɗaya daga cikin shahararrun batutuwa masu ban sha'awa da suka shafi Elbrus, wato yawan aiki. Akwai jita-jita da yawa game da aikin Elbrus, da kuma na iyakacin duniya. Pessimists sun ce yawan amfanin Elbrus yanzu "ba kome ba ne", kuma zai ɗauki shekarun da suka gabata don cim ma masu samar da "manyan" (watau, a halin yanzu, ba). A gefe guda, masu kyakkyawan fata sun ce Elbrus 8C ya riga ya nuna kyakkyawan sakamako, kuma a cikin shekaru biyu masu zuwa, tare da sakin sabbin nau'ikan na'urori masu sarrafawa (Elbrus 16C da 32C), za mu iya "kamawa kuma mu ci gaba" manyan masana'antun sarrafa kayan masarufi na duniya.
Mu a Aerodisk mutane ne masu amfani, don haka mun ɗauki hanya mafi sauƙi kuma mafi fahimta (a gare mu): gwada, rikodin sakamakon kuma kawai zana ƙarshe. A sakamakon haka, mun gudanar da adadi mai yawa na gwaje-gwaje kuma mun gano adadin fasalulluka na aiki na Elbrus 8C e2k architecture (ciki har da masu daɗi) kuma, ba shakka, idan aka kwatanta wannan tare da tsarin ajiya iri ɗaya akan Intel Xeon amd64 na'urori masu sarrafa kayan gini.
Af, za mu yi magana dalla-dalla game da gwaje-gwaje, sakamako da ci gaban tsarin ajiya na gaba akan Elbrus a gidan yanar gizon mu na gaba "OkoloIT" a ranar 15.10.2020 ga Oktoba, 15 a 00: XNUMX. Kuna iya yin rajista ta amfani da hanyar haɗin da ke ƙasa.
→
Gwajin tsayawa
Mun halicci tsayuwa biyu. Dukansu tsaye sun ƙunshi uwar garken da ke aiki da Linux, wanda aka haɗa ta hanyar 16G FC yana canzawa zuwa masu sarrafa ajiya guda biyu, wanda aka shigar da faifai 12 SAS SSD 960 GB (11,5 TB na "ƙarfin albarkatun ƙasa" ko 5,7 TB na ƙarfin "mai amfani", idan muka yi amfani da RAID). -10).
A tsari tsayawa yayi kama da haka.
Tsaya No. 1 e2k (Elbrus)
Tsarin hardware shine kamar haka:
- Sabar Linux (2xIntel Xeon E5-2603 v4 (6 cores, 1,70Ghz), 64 GB DDR4, 2xFC adaftar 16G 2 tashar jiragen ruwa) - 1 pc.
- Canja FC 16 G - 2 inji mai kwakwalwa.
- Tsarin ajiya Aerodisk Vostok 2-E12 (2xElbrus 8C (8 cores, 1,20Ghz), 32 GB DDR3, 2xFE FC-adaftar 16G 2 tashar jiragen ruwa, 12xSAS SSD 960 GB) - 1 pc.
Tsaya No. 2 amd64 (Intel)
Don kwatanta da irin wannan sanyi akan e2k, mun yi amfani da tsarin ajiya mai kama da na'ura mai kama da halaye zuwa amd64:
- Sabar Linux (2xIntel Xeon E5-2603 v4 (6 cores, 1,70Ghz), 64 GB DDR4, 2xFC adaftar 16G 2 tashar jiragen ruwa) - 1 pc.
- Canja FC 16 G - 2 inji mai kwakwalwa.
- Tsarin ajiya Aerodisk Engine N2 (2xIntel Xeon E5-2603 v4 (6 cores, 1,70Ghz), 32 GB DDR4, 2xFE FC-adaftar 16G 2 tashar jiragen ruwa, 12xSAS SSD 960 GB) - 1 pc.
Muhimmiyar sanarwa: Elbrus 8C na'urori masu sarrafawa da aka yi amfani da su a cikin gwajin gwajin DDR3 RAM kawai, wannan ba shakka "mummuna ne, amma ba na dogon lokaci ba." Elbrus 8SV (ba mu da shi a hannun jari tukuna, amma za mu same shi nan ba da jimawa ba) yana goyan bayan DDR4.
Hanyar Gwaji
Don samar da kaya, mun yi amfani da sanannen shirin IO (FIO).
Dukansu tsarin ajiya an saita su bisa ga shawarwarin daidaitawar mu, dangane da buƙatun don babban aiki akan damar toshewa, don haka muna amfani da wuraren waha na faifai na DDP (Dynamic Disk Pool). Don kar a karkatar da sakamakon gwajin, muna kashe matsawa, cirewa da cache RAM akan tsarin ajiya guda biyu.
8 D-LUNs an ƙirƙira su a cikin RAID-10, 500 GB kowanne, tare da jimillar iya aiki na 4 TB (watau kusan 70% na yuwuwar damar amfani da wannan tsarin).
Za a aiwatar da asali da mashahurin yanayin amfani da tsarin ajiya, musamman:
gwaje-gwaje biyu na farko sun yi koyi da aikin DBMS na ma'amala. A cikin wannan rukunin gwaje-gwaje muna sha'awar IOPS da latency.
1) Karatun bazuwar a cikin ƙananan tubalan 4k
a. Girman toshe = 4k
b. Karanta/Rubuta = 100%/0%
c. Yawan aiki = 8
d. Zurfin layi = 32
e. Load hali = Cikakken bazuwar
2) Rikodin bazuwar a cikin ƙananan tubalan 4k
a. Girman toshe = 4k
b. Karanta/Rubuta = 0%/100%
c. Yawan aiki = 8
d. Zurfin layi = 32
e. Load hali = Cikakken bazuwar
gwaje-gwaje biyu na biyu sun yi koyi da aikin sashin nazari na DBMS. A cikin wannan rukunin gwaje-gwajen kuma muna sha'awar IOPS da latency.
3) Karatun jeri a cikin ƙananan tubalan 4k
a. Girman toshe = 4k
b. Karanta/Rubuta = 100%/0%
c. Yawan aiki = 8
d. Zurfin layi = 32
e. Load hali = Jeri
4) Rikodi na jere a cikin ƙananan tubalan 4k
a. Girman toshe = 4k
b. Karanta/Rubuta = 0%/100%
c. Yawan aiki = 8
d. Zurfin layi = 32
e. Load hali = Jeri
Rukuni na uku na gwaje-gwajen sun kwaikwayi aikin karatun yawo (misali: watsa shirye-shiryen kan layi, maido da madogara) da rikodi mai gudana (misali: sa ido na bidiyo, rikodin rikodi). A cikin wannan rukunin gwaje-gwaje, ba mu da sha'awar IOPS, amma a cikin MB/s da kuma latency.
5) Karatun jeri a cikin manyan tubalan 128k
a. Girman toshe = 128k
b. Karanta/Rubuta = 0%/100%
c. Yawan aiki = 8
d. Zurfin layi = 32
e. Load hali = Jeri
6) Rikodi na jeri a cikin manyan tubalan 128k
a. Girman toshe = 128k
b. Karanta/Rubuta = 0%/100%
c. Yawan aiki = 8
d. Zurfin layi = 32
e. Load hali = Jeri
Kowane gwaji zai ɗauki awa ɗaya, ban da lokacin dumama na mintuna 7.
Sakamakon gwaji
An taƙaita sakamakon gwajin a cikin teburi biyu.
Elbrus 8S (SHD Aerodisk Vostok 2-E12)
Intel Xeon E5-2603 v4 (Tsarin Ajiya Aerodisk Engine N2)
Sakamakon ya zama mai ban sha'awa sosai. A cikin lokuta biyu, mun yi amfani da ikon sarrafa tsarin ajiya (70-90% amfani), kuma a cikin wannan yanayin, ribobi da fursunoni na masu sarrafawa biyu sun bayyana a fili.
A cikin duka allunan, gwaje-gwaje inda na'urori masu sarrafawa "ji da kwarin gwiwa" kuma suna nuna sakamako mai kyau ana haskaka su a cikin kore, yayin da yanayin da na'urori "ba sa so" ana haskaka su a cikin orange.
Idan muka yi magana game da bazuwar lodi a cikin ƙananan tubalan, to:
- daga mahangar karatun bazuwar, Intel tabbas yana gaban Elbrus, bambancin shine sau 2;
- daga mahangar yin rikodin bazuwar tabbas zane ne, duka na'urori biyu sun nuna kusan daidai da sakamako mai kyau.
A cikin jerin kaya a cikin ƙananan tubalan hoton ya bambanta:
- duka lokacin karatu da rubutu, Intel yana da mahimmanci (sau 2) a gaban Elbrus. A lokaci guda kuma, idan Elbrus yana da alamar IOPS ƙasa da na Intel, amma yana da kyau (dubu 200-300), to akwai matsala a bayyane tare da jinkiri (sun ninka na Intel sau uku). Kammalawa, sigar Elbrus 8C na yanzu da gaske "ba ya son" lodin jeri a cikin ƙananan tubalan. A bayyane yake akwai wasu ayyuka da za a yi.
Amma a cikin jerin kaya tare da manyan tubalan, hoton daidai yake da akasin haka:
- duka na'urori biyu sun nuna kusan sakamako daidai a MB/s, amma akwai guda ɗaya AMMA.... Ayyukan jinkirin Elbrus shine sau 10 (goma, Karl !!!) mafi kyau (watau ƙasa) fiye da na irin wannan na'ura mai sarrafawa daga Intel (0,4/0,5 ms da 5,1/6,5 ms) . Da farko mun yi tunanin kuskure ne, don haka muka sake duba sakamakon, mun sake gwadawa, amma sake gwadawa ya nuna irin wannan hoto. Wannan babbar fa'ida ce ta Elbrus (da e2k gine-gine gabaɗaya) akan Intel (kuma, saboda haka, gine-ginen amd64). Mu yi fatan wannan nasara ta ci gaba.
Akwai wani fasali mai ban sha'awa na Elbrus, wanda mai karatu mai hankali zai iya kula da shi ta hanyar kallon tebur. Idan ka kalli bambanci tsakanin aikin karantawa da rubutu na Intel, to a duk gwaje-gwajen, karatun yana gaba da rubutu akan matsakaici da kusan 50%+. Wannan shi ne ka'idar da kowa (har da mu) ya saba da shi. Idan ka dubi Elbrus, alamun rubutu sun fi kusa da alamun karatu; karatu yana gaba da rubuce-rubuce, a matsayin mai mulkin, ta 10 - 30%, babu.
Menene ma'anar wannan? Gaskiyar cewa Elbrus "da gaske yana son" rubuce-rubuce, kuma wannan, bi da bi, ya nuna cewa wannan na'ura mai sarrafawa zai zama da amfani sosai a cikin ayyuka inda rubuce-rubucen ya kasance a fili a kan karatun (wanda ya ce dokar Yarovaya?), wanda kuma ba shakka amfani e2k gine, da kuma wannan fa'ida yana buƙatar haɓaka.
Ƙarshe da kuma nan gaba
Gwaje-gwajen kwatankwacin Elbrus da Intel na matsakaicin matsakaici don ayyukan ajiyar bayanai sun nuna kusan daidai kuma daidai da sakamako, yayin da kowane mai sarrafa ya nuna nasa fasali masu ban sha'awa.
Intel ya zarce Elbrus sosai a cikin karatun bazuwar a cikin ƙananan tubalan, da kuma a cikin jerin karatu da rubutu a cikin ƙananan tubalan.
Lokacin rubuta bazuwar cikin ƙananan tubalan, duka na'urori biyu suna nuna sakamako daidai.
Dangane da latency, Elbrus yayi kyau sosai fiye da Intel a cikin kaya mai gudana, watau. a jere karatu da rubutu a cikin manyan tubalan.
Bugu da ƙari, Elbrus, ba kamar Intel ba, yana jurewa daidai da duka karantawa da rubuta lodi, yayin da tare da Intel, karatun koyaushe ya fi rubutu.
Dangane da sakamakon da aka samu, za mu iya zana ƙarshe game da aiwatar da tsarin ajiyar bayanan Aerodisk Vostok akan processor Elbrus 8C a cikin ayyuka masu zuwa:
- tsarin bayanai tare da fifikon ayyukan rubuce-rubuce;
- samun damar fayil;
- watsa shirye-shiryen kan layi;
- CCTV;
- madadin;
- abubuwan watsa labarai.
Ƙungiyar MCST har yanzu tana da wani abu don yin aiki a kai, amma sakamakon aikin su ya riga ya bayyana, wanda, ba shakka, ba zai iya yin farin ciki ba.
An gudanar da waɗannan gwaje-gwaje akan Linux kernel don nau'in e2k 4.19; a halin yanzu a cikin gwajin beta (a cikin MCST, a cikin Basalt SPO, da kuma nan a cikin Aerodisk) akwai Linux kernel 5.4-e2k, wanda, a tsakanin sauran abubuwa, yana da An sabunta tsarin tsarawa sosai da haɓakawa da yawa don ingantattun abubuwan tafiyar da jihar mai sauri. Hakanan, musamman don kernels na reshen 5.x.x, MCST JSC yana fitar da sabon mai tara LCC, sigar 1.25. Dangane da sakamakon farko, akan wannan na'ura ta Elbrus 8C, wani sabon kwaya wanda aka haɗa tare da sabon mai tarawa, yanayin kwaya, kayan aikin tsarin da ɗakunan karatu kuma, a zahiri, software na Aerodisk VOSTOK zai ba da damar haɓaka haɓaka mai mahimmanci. Kuma wannan ba tare da maye gurbin kayan aiki ba - akan na'ura guda ɗaya kuma tare da mitoci iri ɗaya.
Muna sa ran fitowar wani nau'in Aerodisk VOSTOK dangane da kernel 5.4 zuwa ƙarshen shekara, kuma da zaran an kammala aikin sabon sigar, za mu sabunta sakamakon gwajin kuma mu buga su anan.
Idan muka koma farkon labarin kuma mu amsa tambayar, wane ne daidai: ssimists suka ce Elbrus "ba kome ba" da kuma ba za su taba cim tare da manyan processor masana'antun, ko optimists suka ce "sun riga sun kusan kama. tashi kuma zai wuce"? Idan muka ci gaba ba daga ra'ayi da ra'ayi na addini ba, amma daga gwaje-gwaje na gaske, to lallai masu fata suna da gaskiya.
Elbrus ya riga ya nuna kyakkyawan sakamako idan aka kwatanta da na'urori masu sarrafawa na amd64 na tsakiyar matakin. Elbrus 8-ke, ba shakka, yana da nisa daga saman-na-layi na masu sarrafa sabar sabar daga Intel ko AMD, amma ba a yi niyya a wurin ba; za a fitar da masu sarrafawa 16C da 32C don wannan dalili. Sannan zamuyi magana.
Mun fahimci cewa bayan wannan labarin za a sami ƙarin tambayoyi game da Elbrus, don haka mun yanke shawarar tsara wani webinar kan layi "OkoloIT" don amsa waɗannan tambayoyin kai tsaye.
A wannan lokacin baƙonmu zai kasance Mataimakin Babban Darakta na kamfanin MCST, Konstantin Trushkin. Kuna iya yin rajista don webinar ta amfani da hanyar haɗin da ke ƙasa.
→
Na gode duka, kamar koyaushe, muna sa ido ga zargi mai ma'ana da tambayoyi masu ban sha'awa.
source: www.habr.com