Injiniyoyin Facebook (an dakatar da Tarayyar Rasha) sun buga rahoto game da aiwatar da fasahar TMO (Transparent Memory Offloading) a bara, wanda ke ba da damar tanadi mai yawa a cikin RAM akan sabobin ta hanyar kawar da bayanan sakandaren da ba a buƙata don aiki zuwa masu rahusa, kamar NVMe. SSD - faifai. Facebook yayi kiyasin cewa yin amfani da TMO na iya adana 20 zuwa 32% na RAM akan kowace uwar garken. An tsara maganin don amfani a cikin abubuwan more rayuwa inda aikace-aikacen ke gudana a cikin keɓaɓɓen kwantena. Abubuwan da ke gefen kernel na TMO an riga an haɗa su a cikin kernel na Linux.
A gefen kernel na Linux, fasahar tana samun goyan bayan tsarin tsarin PSI (Matsa lamba Stall Information), ana samun farawa tare da sakin 4.20. An riga an yi amfani da PSI a cikin masu sarrafa ƙananan ƙwaƙwalwar ajiya daban-daban kuma yana ba ku damar bincika bayanai game da lokacin jira don samun albarkatu daban-daban (CPU, ƙwaƙwalwar ajiya, I/O). Tare da PSI, masu sarrafa sararin samaniya na masu amfani zasu iya tantance matakan nauyin tsarin daidai da tsarin tafiyar hawainiya, ba da damar gano abubuwan da ba a sani ba da wuri, kafin su sami tasiri mai tasiri akan aiki.
A cikin sararin mai amfani, ana samar da TMO ta bangaren Senpai, wanda, ta hanyar cgroup2, yana daidaita iyakar ƙwaƙwalwar ajiya don kwantena aikace-aikacen dangane da bayanan da aka karɓa daga PSI. Senpai yayi nazarin alamun farkon ƙarancin albarkatu ta hanyar PSI, yana kimanta ƙwarewar aikace-aikacen zuwa raguwar samun damar ƙwaƙwalwar ajiya kuma yayi ƙoƙarin tantance ƙaramin girman ƙwaƙwalwar da ake buƙata ta akwati, wanda bayanan da ake buƙata don aiki ya kasance a cikin RAM, da rakiyar. bayanan da aka daidaita a cikin cache fayil ko ba a yi amfani da su kai tsaye ba a yanzu, an tilasta su zuwa ɓangaren musanyawa.
Don haka, jigon TMO shine kiyaye matakai akan ingantaccen abinci dangane da amfani da ƙwaƙwalwar ajiya, tilasta yin musanyawa na shafukan ƙwaƙwalwar da ba a yi amfani da su ba waɗanda korarsu ba ta da tasiri sosai akan aikin (misali, shafuka tare da lambar da aka yi amfani da su kawai yayin farawa, da amfani guda ɗaya). bayanai a cikin cache diski). Ba kamar fitar da bayanai zuwa ɓangaren musanya ba don amsa matsa lamba na ƙwaƙwalwar ajiya, a cikin bayanan TMO ana fitar da su bisa ga hasashe mai ƙarfi.
Ɗaya daga cikin ma'auni don fitarwa shine rashin samun dama ga shafin ƙwaƙwalwar ajiya na minti 5. Irin waɗannan shafukan ana kiran su shafukan ƙwaƙwalwar ajiyar sanyi kuma a matsakaita suna da kusan kashi 35% na ƙwaƙwalwar aikace-aikacen (ya danganta da nau'in aikace-aikacen, akwai kewayon daga 19% zuwa 65%). Preemption yana la'akari da ayyukan da ke da alaƙa da shafukan ƙwaƙƙwaran da ba a san su ba (ƙwaƙwalwar ajiyar da aikace-aikacen ya keɓe) da ƙwaƙwalwar ajiyar da aka yi amfani da shi don caching fayil (wanda kernel ke warewa). A wasu aikace-aikacen babban amfani shine ƙwaƙwalwar ajiyar da ba a san su ba, amma a wasu kuma cache fayil ɗin yana da mahimmanci. Don guje wa rashin daidaiton korar cache, TMO yana amfani da sabon tsarin bincike wanda ke korar shafukan da ba a san su ba da shafukan da ke da alaƙa da cache fayil daidai gwargwado.
Tura shafukan da ba safai ake amfani da su ba cikin ƙwaƙwalwar ajiyar hankali baya da tasiri sosai akan aiki, amma yana iya rage tsadar kayan masarufi. Ana juyar da bayanai zuwa faifan SSD ko zuwa wurin musanyawa da aka matsa a cikin RAM. Dangane da tsadar adana bayanan, amfani da NVMe SSD ya kai sau 10 mai rahusa fiye da amfani da matsawa a cikin RAM.
source: budenet.ru