Idan kuna gudanar da kayan aikin kama-da-wane bisa VMware vSphere (ko duk wani tarin fasaha), mai yiwuwa kuna yawan jin koke-koke daga masu amfani: "Na'urar tana jinkirin!" A cikin wannan jerin labaran zan yi nazarin ma'auni na aiki kuma in gaya muku abin da kuma dalilin da ya sa ya ragu da kuma yadda za a tabbatar da cewa ba ya raguwa.
Zan yi la'akari da abubuwa masu zuwa na aikin injin kama-da-wane:
- CPU,
- SAURARA,
- DISK,
- Network.
Zan fara da CPU.
Don nazarin aikin za mu buƙaci:
- vCenter Performance Counters - ƙididdiga masu aiki, waɗanda za a iya kallon jadawali ta hanyar Client vSphere. Ana samun bayanai akan waɗannan ƙididdiga a cikin kowane sigar abokin ciniki (abokin ciniki "kauri" a cikin C #, abokin ciniki na yanar gizo a cikin Flex da abokin ciniki na yanar gizo a cikin HTML5). A cikin waɗannan labaran za mu yi amfani da hotunan kariyar kwamfuta daga abokin ciniki na C #, kawai saboda sun fi kyau a cikin ƙananan :)
- ESXTOP - mai amfani da ke gudana daga layin umarni na ESXi. Tare da taimakonsa, zaku iya samun ƙimar ƙididdiga masu aiki a cikin ainihin lokaci ko loda waɗannan ƙimar na wani ɗan lokaci cikin fayil .csv don ƙarin bincike. Na gaba, zan ba ku ƙarin bayani game da wannan kayan aiki kuma in samar da hanyoyi masu amfani da yawa zuwa takardu da labarai kan batun.
A bit of ka'idar
A cikin ESXi, wani tsari na daban - duniya a cikin kalmomin VMware - yana da alhakin aiki na kowane vCPU (cibiyar na'ura mai kama-da-wane). Hakanan akwai hanyoyin sabis, amma daga ra'ayi na nazarin ayyukan VM ba su da ban sha'awa.
Tsari a cikin ESXi na iya kasancewa cikin ɗayan jihohi huɗu:
- Run - tsarin yana yin wasu ayyuka masu amfani.
- Jira - tsarin ba ya yin wani aiki (rago) ko yana jiran shigarwa / fitarwa.
- Farashin - yanayin da ke faruwa a cikin injunan kama-da-wane. Yana faruwa a lokacin da mai tsara tsarin CPU na hypervisor (ESXi CPU Scheduler) ba zai iya tsara aiwatar da aiwatar da lokaci guda na duk nau'ikan injunan na'ura mai aiki akan sabar sabar ta zahiri ba. A cikin duniyar zahiri, duk kayan aikin sarrafawa suna aiki a layi daya, OS baƙon da ke cikin VM yana tsammanin irin wannan hali, don haka hypervisor dole ne ya rage abubuwan VM waɗanda ke da ikon gama zagayen agogon su cikin sauri. A cikin nau'ikan ESXi na zamani, mai tsara tsarin CPU yana amfani da wata hanyar da ake kira tsarin haɗin gwiwa mai annashuwa: hypervisor yana la'akari da tazarar da ke tsakanin "mafi sauri" da "mafi saurin" inji mai mahimmanci (skew). Idan tazarar ta wuce ƙayyadaddun ƙofa, core mai sauri yana shiga cikin halin da ake ciki. Idan muryoyin VM suna ciyar da lokaci mai yawa a cikin wannan jihar, zai iya haifar da al'amuran aiki.
- Ready - tsarin yana shiga cikin wannan yanayin lokacin da hypervisor ya kasa ware albarkatun don aiwatar da shi. Babban shirye-shiryen dabi'u na iya haifar da matsalolin aikin VM.
Na'ura mai mahimmanci na kayan aikin CPU
Amfanin CPU, %. Yana nuna adadin yawan amfanin CPU na wani ɗan lokaci.
Yadda za a tantance? Idan VM koyaushe yana amfani da CPU a 90% ko kuma akwai kololuwa har zuwa 100%, to muna da matsaloli. Ana iya bayyana matsalolin ba kawai a cikin aikin "jinkirin" na aikace-aikacen a cikin VM ba, amma har ma a cikin rashin isa ga VM akan hanyar sadarwa. Idan tsarin sa ido ya nuna cewa VM lokaci-lokaci yana faɗuwa, kula da kololuwa a cikin jadawali Amfanin CPU.
Akwai daidaitaccen ƙararrawa wanda ke nuna nauyin CPU na injin kama-da-wane:
Abin da ya yi? Idan VM's CPU Usage yana ci gaba da tafiya ta cikin rufin, to zaku iya yin tunani game da haɓaka adadin vCPUs (abin takaici, wannan baya taimakawa koyaushe) ko matsar da VM zuwa sabar tare da masu sarrafawa masu ƙarfi.
Amfani da CPU a cikin MHz
A cikin zane-zane akan amfani da VCETREler a cikin% Zaka gani kawai ga dukkan na'urori masu amfani; babu wasu zane-zane don ka'idodi na mutum. Ga kowane cibiya zaka iya ganin Amfani a cikin MHz.
Yadda za a tantance? Yana faruwa cewa aikace-aikacen ba a inganta shi don gine-ginen multi-core: yana amfani da cibiya ɗaya kawai 100%, sauran kuma ba su da nauyi. Misali, tare da saitunan madadin tsoho, MS SQL yana fara aiwatarwa akan cibiya ɗaya kawai. Sakamakon haka, ajiyar ajiyar yana raguwa ba saboda jinkirin saurin diski ba (wannan shine abin da mai amfani ya fara koka game da shi), amma saboda processor ba zai iya jurewa ba. An warware matsalar ta hanyar canza sigogi: madadin ya fara aiki a layi daya a cikin fayiloli da yawa (bi da bi, a cikin matakai da yawa).
Misali na nauyin da bai dace ba akan murhu.
Akwai kuma wani yanayi (kamar yadda yake a cikin jadawali na sama) lokacin da aka ɗora nauyin muryoyin ba daidai ba kuma wasu daga cikinsu suna da kololuwa na 100%. Kamar yadda tare da lodawa guda ɗaya kawai, ƙararrawa don Amfani da CPU ba zai yi aiki ba (na duka VM ne), amma za a sami matsalolin aiki.
Abin da ya yi? Idan software ɗin da ke cikin injin kama-da-wane ta ɗora nauyin maƙallan ba daidai ba (yana amfani da cibiya ɗaya kawai ko wani ɓangare na cores), babu wani amfani wajen ƙara yawan su. A wannan yanayin, yana da kyau a matsar da VM zuwa uwar garken tare da masu sarrafawa masu ƙarfi.
Hakanan zaka iya gwada duba saitunan amfani da wutar lantarki a cikin uwar garken BIOS. Yawancin masu gudanar da gudanarwa suna ba da damar Yanayin Babban Aiki a cikin BIOS kuma ta haka ne ke kashe jihohin C da P-jihohin fasahar ceton makamashi. Na'urori na zamani na Intel suna amfani da fasahar Turbo Boost, wanda ke ƙara yawan adadin nau'in sarrafawa guda ɗaya tare da kuɗin wasu nau'in. Amma yana aiki ne kawai lokacin da aka kunna fasahar ceton makamashi. Idan muka kashe su, na'ura mai sarrafawa ba zai iya rage yawan wutar lantarkin da ba a ɗorawa ba.
VMware yana ba da shawarar kar a kashe fasahohin ceton wuta akan sabar, amma zabar hanyoyin da ke barin sarrafa wutar lantarki zuwa ga hypervisor gwargwadon yiwuwa. A wannan yanayin, kuna buƙatar zaɓar Babban Ayyuka a cikin saitunan amfani da wutar lantarki na hypervisor.
Idan kuna da nau'ikan VM guda ɗaya (ko VM) a cikin kayan aikin ku waɗanda ke buƙatar haɓaka mitar CPU, daidaitaccen amfani da wutar lantarki na iya haɓaka aikinsu sosai.
CPU Shirye
Idan VM core (vCPU) yana cikin Yanayin Shirye, baya yin aiki mai amfani. Wannan yanayin yana faruwa lokacin da hypervisor bai sami ainihin ainihin zahiri na kyauta wanda za'a iya sanya tsarin vCPU na injin kama-da-wane.
Yadda za a tantance? Yawanci, idan muryoyin injin kama-da-wane suna cikin jihar Shirye fiye da kashi 10% na lokaci, zaku lura da al'amuran aiki. A taƙaice, fiye da 10% na lokacin VM yana jiran albarkatun jiki don samuwa.
A cikin vCenter zaku iya duba ƙididdiga guda 2 masu alaƙa da Shiryewar CPU:
- shiri,
- Shirya.
Ana iya duba ƙimar ƙididdiga biyu ga duka VM da kuma ga kowane nau'i.
Shirye-shiryen yana nuna ƙimar nan da nan azaman kashi, amma kawai a cikin Real-lokaci (bayanai na awa na ƙarshe, tazarar ma'auni 20 seconds). Zai fi kyau a yi amfani da wannan ƙididdiga kawai don bincika matsalolin "zafi a kan dugadugan".
Hakanan za'a iya duba ƙimar ƙima mai ƙima daga hangen nesa na tarihi. Wannan yana da amfani don kafa alamu da kuma zurfin nazarin matsalar. Misali, idan na'urar kama-da-wane ta fara fuskantar matsalolin aiki a wani ɗan lokaci, zaku iya kwatanta tazara na ƙimar CPU Ready tare da jimlar kaya akan uwar garken inda wannan VM ke gudana, kuma ɗaukar matakan rage nauyin (idan DRS) kasa).
Shirye, ba kamar Shirye ba, ana nunawa ba a cikin kashi ɗaya ba, amma a cikin millise seconds. Wannan nau'in ƙira ne na Taƙaitawa, wato, yana nuna tsawon lokacin auna ma'aunin VM core yana cikin jihar Shirye. Kuna iya canza wannan ƙimar zuwa kashi ta amfani da dabara mai sauƙi:
(CPU shirye-shiryen taƙaita ƙimar / (tsarar da tazarar sabunta tazara a cikin daƙiƙa * 1000)) * 100 = CPU shirye %
Misali, don VM a cikin jadawali da ke ƙasa, ƙimar da aka shirya don duk injin kama-da-wane zai kasance kamar haka:
Lokacin ƙididdige adadin Ready, ya kamata ku kula da maki biyu:
- Ƙimar da aka Shirya a duk VM ita ce jimlar Shirye a kan murhu.
- Tazarar aunawa. Don ainihin lokacin yana da daƙiƙa 20, kuma, alal misali, akan jadawalin yau da kullun yana da daƙiƙa 300.
Tare da magance matsalar aiki, ana iya rasa waɗannan mahimman bayanai cikin sauƙi kuma ana iya ɓata lokaci mai mahimmanci akan magance matsalolin da ba su wanzu.
Bari mu lissafta Ready bisa bayanai daga jadawali da ke ƙasa. (324474/(20*1000))*100 = 1622% ga dukkan VM. Idan ka kalli muryoyin ba abin ban tsoro ba ne: 1622/64 = 25% kowace sahihanci. A wannan yanayin, kama yana da sauƙin gano: ƙimar Ready ba ta gaskiya ba ce. Amma idan muna magana game da 10-20% ga dukan VM tare da nau'i-nau'i da yawa, to ga kowane mahimmancin ƙima na iya kasancewa cikin kewayon al'ada.
Abin da ya yi? Ƙimar Shirye mai girma tana nuna cewa uwar garken ba ta da isassun albarkatun sarrafawa don aiki na yau da kullun na injuna. A irin wannan yanayi, abin da ya rage shi ne a rage yawan yin rajista ta hanyar sarrafawa (vCPU:pCPU). Babu shakka, ana iya samun wannan ta hanyar rage ma'auni na VMs ɗin da ke akwai ko ta ƙaura wani ɓangare na VMs zuwa wasu sabar.
Tsayawa tare
Yadda za a tantance? Wannan ma'aunin ma na nau'in Takaitawa ne kuma ana jujjuya shi zuwa kashi-kashi kamar yadda Ready:
(CPU co-stop summming value / (tsarin sabuntawa ta tsoho a cikin dakika * 1000)) * 100 = CPU co-stop %
Anan kuna buƙatar kula da adadin muryoyin akan VM da tazarar ma'auni.
A cikin jihar costop, kernel baya yin aiki mai amfani. Tare da madaidaicin zaɓi na girman VM da nauyin al'ada akan uwar garken, ma'aunin tasha ya kamata ya kasance kusa da sifili.
A wannan yanayin, kaya a fili ba daidai ba ne :)
Abin da ya yi? Idan VM da yawa tare da adadi mai yawa na cores suna gudana akan hypervisor ɗaya kuma akwai oversubscription akan CPU, to counter-stop counter na iya ƙaruwa, wanda zai haifar da matsaloli tare da aikin waɗannan VMs.
Hakanan, haɗin gwiwa zai ƙaru idan muryoyin VM guda ɗaya suna amfani da zaren akan ainihin uwar garken jiki ɗaya tare da kunna hyper-treading. Wannan yanayin na iya tasowa, alal misali, idan VM yana da ƙarin ƙira fiye da samuwa a kan uwar garke inda yake gudana, ko kuma idan an kunna saitin "preferHT" don VM. Kuna iya karanta game da wannan saitin .
Don guje wa matsaloli tare da aikin VM saboda babban haɗin gwiwa, zaɓi girman VM daidai da shawarwarin masana'antun software da ke aiki akan wannan VM da kuma damar uwar garken jiki inda VM ke gudana.
Kada a ƙara cores a ajiye; wannan na iya haifar da matsalolin aiki ba kawai ga VM kanta ba, har ma da maƙwabta akan sabar.
Sauran ma'aunin CPU masu amfani
Run - nawa lokaci (ms) yayin lokacin auna vCPU yana cikin yanayin RUN, wato, yana aiwatar da aiki mai amfani.
malalaci - tsawon lokacin (ms) yayin lokacin awo vCPU tana cikin yanayin rashin aiki. Maɗaukakin ƙimar rashin aiki ba matsala ba ne, vCPU kawai ba shi da "abin da za a yi."
Jira - tsawon lokacin (ms) yayin lokacin auna vCPU yana cikin yanayin Jira. Tunda an haɗa IDLE a cikin wannan ma'aunin, ƙimar jira mai girma suma baya nuna matsala. Amma idan Jira IDLE ya yi ƙasa sosai lokacin jira yana da girma, yana nufin VM yana jiran ayyukan I/O don kammalawa, kuma wannan, bi da bi, na iya nuna matsala game da aikin rumbun kwamfutarka ko kowane na'ura mai kama da VM.
Max Limited - tsawon lokacin (ms) a lokacin aunawa vCPU ta kasance a cikin Yanayin Shirye saboda ƙayyadaddun iyaka na albarkatu. Idan aikin yana da ƙarancin fa'ida, to yana da amfani a duba ƙimar wannan counter da iyakar CPU a cikin saitunan VM. VMs tabbas suna da iyakoki waɗanda ba ku sani ba. Misali, wannan yana faruwa lokacin da aka rufe VM daga samfuri wanda aka saita iyakar CPU akansa.
Sauya jira - tsawon lokacin aunawa vCPU ya jira aiki tare da VMkernel Swap. Idan dabi'un wannan counter sun fi sifili, to tabbas VM yana da matsalolin aiki. Za mu yi magana game da SWAP a cikin labarin game da masu lissafin RAM.
ESXTOP
Idan ƙididdiga masu aiki a cikin vCenter suna da kyau don nazarin bayanan tarihi, to, nazarin aiki na matsalar ya fi kyau a yi a ESXTOP. Anan, ana gabatar da duk ƙimar a cikin shirye-shiryen da aka yi (babu buƙatar fassara wani abu), kuma mafi ƙarancin lokacin awo shine 2 seconds.
Ana kiran allon ESXTOP na CPU tare da maɓallin "c" kuma yayi kama da wannan:
Don saukakawa, zaku iya barin tsarin injin kama-da-wane kawai ta latsa Shift-V.
Don duba ma'auni don nau'ikan nau'ikan VM guda ɗaya, danna "e" kuma shigar da GID na VM na sha'awa (30919 a cikin hoton da ke ƙasa):
Bari in taƙaice ta cikin ginshiƙan da aka gabatar ta hanyar tsoho. Ana iya ƙara ƙarin ginshiƙai ta latsa "f".
NWLD (Yawan Duniya) - adadin matakai a cikin rukuni. Don faɗaɗa ƙungiyar kuma duba ma'auni na kowane tsari (misali, ga kowane cibiya a cikin VM mai yawan gaske), danna "e". Idan akwai tsari fiye da ɗaya a cikin rukuni, to, ƙimar awo na ƙungiyar daidai yake da jimillar ma'auni na matakan mutum ɗaya.
% AMFANI – nawa uwar garken CPU hawan keke ake amfani da wani tsari ko rukuni na matakai.
GUDU - tsawon lokacin lokacin auna tsarin yana cikin yanayin RUN, watau. ya yi aiki mai amfani. Ya bambanta da %USED a cikin cewa baya la'akari da hyper-threading, mita da kuma lokacin da aka kashe akan ayyukan tsarin (% SYS).
% SYS - lokacin da aka kashe akan ayyukan tsarin, misali: katse aiki, I / O, aikin cibiyar sadarwa, da dai sauransu. Darajar na iya zama babba idan VM yana da babban I / O.
% OVRLP - nawa lokacin jigon jiki wanda tsarin VM ke gudana akan ayyukan wasu matakai.
Waɗannan ma'auni suna da alaƙa da juna kamar haka:
% AMFANI = % GUDU + % SYS - % OVRLP.
Yawanci ma'aunin %USED yana da ƙarin bayani.
% JIRA - tsawon lokacin lokacin auna tsarin yana cikin yanayin Jira. Yana kunna IDLE.
%IDLE - tsawon lokacin lokacin auna tsarin yana cikin jihar IDLE.
% SWPWT - tsawon lokacin aunawa vCPU ya jira aiki tare da VMkernel Swap.
%VMWAIT - tsawon lokacin aunawa vCPU tana cikin yanayin jiran wani taron (yawanci I/O). Babu madaidaicin ma'auni a vCenter. Babban dabi'u suna nuna matsaloli tare da I/O akan VM.
% JIRA = %VMWAIT + %IDLE + %SWPWT.
Idan VM baya amfani da VMkernel Swap, to, lokacin da ake nazarin matsalolin aiki yana da kyau a duba %VMWAIT, tunda wannan ma'aunin baya la'akari da lokacin da VM ba ta yin komai (%IDLE).
% RDY - tsawon lokacin lokacin auna tsarin ya kasance a cikin jihar Shirye.
%CSTP - tsawon lokacin lokacin ma'auni tsarin ya kasance a cikin jihar costop.
%MLTD - tsawon lokacin aunawa vCPU ta kasance a cikin Yanayin Shirye saboda ƙayyadaddun iyaka na albarkatu.
% JIRA + % RDY + % CSTP + % RUN = 100% – ainihin VM koyaushe yana cikin ɗayan waɗannan jihohi huɗu.
CPU akan hypervisor
Har ila yau vCenter yana da ƙididdigar aikin CPU don hypervisor, amma ba su da wani abin ban sha'awa - su ne kawai jimlar ƙididdiga don duk VMs akan sabar.
Hanya mafi dacewa don duba matsayin CPU akan uwar garken shine akan Takaitaccen shafin:
Ga uwar garken, da kuma na injin kama-da-wane, akwai madaidaicin ƙararrawa:
Lokacin da nauyin uwar garken CPU ya yi girma, VMs da ke gudana a kai suna fara fuskantar matsalolin aiki.
A cikin ESXTOP, ana gabatar da bayanan lodin CPU na uwar garken a saman allon. Baya ga daidaitaccen nauyin CPU, wanda ba shi da cikakken bayani ga masu haɓakawa, akwai ƙarin ma'auni guda uku:
CORE UTIL(%) – loading da zahiri uwar garken core. Wannan ma'aunin yana nuna adadin lokacin da ainihin ya yi aikin yayin lokacin aunawa.
PCPU UTIL(%) - idan an kunna hyper-threading, to akwai zaren guda biyu (PCPU) kowane ainihin zahiri. Wannan awo yana nuna tsawon lokacin da kowane zaren ya ɗauka don kammala aiki.
PCPU AMFANI(%) - daidai da PCPU UTIL (%), amma yana la'akari da ƙimar mitar mita (ko dai rage yawan mitar don dalilai na ceton makamashi, ko ƙara yawan mitar saboda fasahar Turbo Boost) da hyper-threading.
PCPU_USED% = PCPU_UTIL% * ingantaccen mitar core / nominal core mita.
A cikin wannan hoton hoton, don wasu muryoyin, saboda Turbo Boost, ƙimar USED ta fi 100%, tunda ainihin mitar ta fi na ƙima.
Kalmomi kaɗan game da yadda ake ɗaukar hyper-threading. Idan an aiwatar da matakai 100% na lokaci akan zaren guda biyu na ainihin uwar garken, yayin da ainihin ke aiki a mitar ƙididdiga, to:
- CORE UTIL na ainihin zai zama 100%,
- PCPU UTIL na zaren biyu zai zama 100%,
- PCPU AMFANI da zaren biyu zai zama 50%.
Idan duka zaren ba su yi aiki ba 100% na lokacin a lokacin aunawa, to, a cikin waɗancan lokutan lokacin da zaren ke aiki a layi daya, PCPU AMFANI da muryoyin ana raba kashi biyu.
ESXTOP kuma yana da allo tare da sigogin amfani da wutar lantarki na CPU. Anan zaka iya ganin ko uwar garken yana amfani da fasahar ceton makamashi: C-states da P-states. An kira shi da maɓallin "p":
Matsalolin Ayyukan CPU gama gari
A ƙarshe, zan wuce abubuwan da ke haifar da matsaloli tare da aikin VM CPU kuma in ba da gajerun shawarwari don warware su:
Gudun agogon tsakiya bai isa ba. Idan ba zai yiwu ba don haɓaka VM ɗin ku zuwa mafi ƙarfin murhu, zaku iya gwada canza saitunan wutar lantarki don sa Turbo Boost yayi aiki sosai.
Ƙimar VM mara daidai (yawanci da yawa). Idan kun shigar da ƴan ƙira, za a sami babban nauyin CPU akan VM. Idan akwai da yawa, kama babban haɗin gwiwa.
Babban oversubscribe na CPU akan uwar garken. Idan VM yana da Babban Shirye, rage yawan biyan kuɗin CPU.
NUMA topology mara daidai akan manyan VMs. NUMA topology gani da VM (vNUMA) dole ne ya dace da NUMA topology na uwar garken (pNUMA). An rubuta bincike da hanyoyin magance wannan matsala, alal misali, a cikin littafin . Idan ba kwa son zurfafa zurfafa kuma ba ku da hani na lasisi akan OS ɗin da aka shigar akan VM, yi kwasfa masu kama da yawa akan VM, cibiya ɗaya a lokaci guda. Ba za ku yi asara da yawa ba :)
Wannan ke gare ni game da CPU. Yi tambayoyi. A kashi na gaba zan yi magana game da RAM.
hanyoyi masu amfani
source: www.habr.com