Ukuthuthukiswa kwe-Btrfs kuhlongozwa ukuthi kufakwe ku-Linux 6.2 kernel ukuze kulungiswe inkinga "yembobo yokubhala" ekusetshenzisweni kwe-RAID 5/6. Ingqikithi yenkinga ikhuphukela eqinisweni lokuthi uma kwenzeka ukuphahlazeka ngesikhathi sokurekhoda, ekuqaleni akunakwenzeka ukuqonda ukuthi iyiphi ibhulokhi ukuthi yimaphi amadivayisi e-RAID abhalwe ngendlela efanele, futhi lapho ukurekhoda kungaqedwanga. Uma uzama ukwakha kabusha i-RAID kulesi simo, amabhulokhi ahambisana namabhulokhi abhalwe phansi angase onakaliswe ngoba isimo samabhulokhi e-RAID asivunyelanisiwe. Le nkinga yenzeka kunoma yiziphi izinhlelo ze-RAID1/5/6 lapho izinyathelo ezikhethekile zingathathwanga ukulwa nalo mphumela.
Ekusetshenzisweni kwe-RAID, njenge-RAID1 kuma-btrfs, le nkinga ixazululwa ngokusebenzisa ama-checksums kuwo womabili amakhophi, uma kukhona ukungafani, idatha ivele ibuyiselwe kusuka kukhophi yesibili. Le ndlela iphinde isebenze uma idivayisi ethile iqala ukunikeza idatha engalungile esikhundleni sokwehluleka okuphelele.
Kodwa-ke, esimweni se-RAID5/6, uhlelo lwefayela alugcini ama-checksums amabhulokhi wokulinganisa: esimweni esivamile, ukufaneleka kwamabhulokhi kuhlolwa ukuthi wonke afakwe i-checksum, futhi i-parity block ingakwazi. kudalwe kabusha kusuka kudatha. Nokho, endabeni yokurekhoda ingxenye, le ndlela ingase ingasebenzi ezimweni ezithile. Kulesi simo, lapho ubuyisela uhlu, kungenzeka ukuthi amabhlogo awela ngaphansi kwerekhodi elingaphelele azobuyiselwa ngokungalungile.
Endabeni yama-btrfs, le nkinga ifaneleka kakhulu uma ukubhala okukhiqizwayo kukuncane kunomugqa. Kulokhu, isistimu yefayela kufanele yenze umsebenzi wokufunda-modify-write (RMW). Uma ihlangabezana namabhulokhi abhalwayo asaqhubekayo, khona-ke ukusebenza kwe-RMW kungase kubangele ukonakala okungeke kutholwe, kungakhathaliseki ukuthi yiziphi izimali zokuhlola. Abathuthukisi benze izinguquko lapho ukusebenza kwe-RMW kuhlola isheke lamabhulokhi ngaphambi kokwenza lo msebenzi, futhi uma kunesidingo, ukutholwa kwedatha nakho kwenza isheke lesheke ngemva kokubhala. Ngeshwa, esimweni sokubhala umugqa ongaphelele (RMW), lokhu kuholela ku-overhead eyengeziwe yokubala amasheke, kodwa kukhulisa kakhulu ukwethembeka. Ku-RAID6, i-logic enjalo ayikakalungi, nokho, ukwehluleka okunjalo ku-RAID6, kuyadingeka ukuthi ukubhala kuhluleke kumadivayisi angu-2 ngesikhathi esisodwa, okungenzeka mancane.
Ukwengeza, singaqaphela izincomo ngokusetshenziswa kwe-RAID5 / 6 konjiniyela, ingqikithi yayo ukuthi kuma-Btrfs iphrofayela yokugcina imethadatha nedatha ingase ihluke. Kulokhu, ungasebenzisa iphrofayela ye-RAID1 (isibuko) noma i-RAID1C3 (amakhophi angu-3) ukuze uthole imethadatha, kanye ne-RAID5 noma i-RAID6 ngedatha. Lokhu kuqinisekisa ukuvikelwa okuthembekile kwemethadatha kanye nokungabikho "komgodi wokubhala", ngakolunye uhlangothi, kanye nokusetshenziswa kahle kwesikhala, okujwayelekile ku-RAID5/6, ngakolunye. Lokhu kugwema ukonakala kumethadatha, futhi ukonakala kwedatha kungalungiswa.
Kungaphawulwa futhi ukuthi kuma-SSD aku-Btrfs ku-6.2 kernel, ukukhishwa okuvumelanayo kokusebenza kokuthi "lahla" (ukumaka amabhlogo akhululiwe angeke esagcinwa ngokomzimba) kuzokwenziwa kusebenze ngokuzenzakalelayo. Inzuzo yale modi ukusebenza okuphezulu ngenxa yeqembu eliphumelelayo lemisebenzi "yokulahla" kulayini kanye nokucutshungulwa okuqhubekayo komugqa yisiphathi sangemuva, ngenxa yokuthi imisebenzi evamile ye-FS ayinciphisi, njengoba kwenzeka ngokuvumelanisa " lahla" njengoba amabhlogo ekhululwa, futhi i-SSD ingenza izinqumo ezingcono. Ngakolunye uhlangothi, ngeke usadinga ukusebenzisa izinsiza ezifana ne-fstrim, njengoba wonke amabhulokhi atholakalayo azosulwa ku-FS ngaphandle kwesidingo sokuskena okwengeziwe futhi ngaphandle kokunciphisa ukusebenza.
Source: opennet.ru