{"id":113046,"date":"2024-01-22T11:28:02","date_gmt":"2024-01-22T09:28:02","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/dlya-yadra-linux-podgotovleny-optimizaczii-povyshayushhie-proizvoditelnost-planirovshhikov-vvoda-vyvoda"},"modified":"2024-01-22T11:28:02","modified_gmt":"2024-01-22T09:28:02","slug":"dlya-yadra-linux-podgotovleny-optimizaczii-povyshayushhie-proizvoditelnost-planirovshhikov-vvoda-vyvoda","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sq\/blog\/news\/dlya-yadra-linux-podgotovleny-optimizaczii-povyshayushhie-proizvoditelnost-planirovshhikov-vvoda-vyvoda","title":{"rendered":"P\u00ebr kernelin Linux jan\u00eb p\u00ebrgatitur optimizime q\u00eb rrisin performanc\u00ebn e planifikuesve t\u00eb input\/output.","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
Jens Axboe, krijuesi i io_uring dhe planifikuesve t\u00eb hyrjes\/dalitjes CFQ, Deadline dhe Noop, vazhdoi eksperimentet e tij me optimizimin e hyrjes\/dalitjes n\u00eb b\u00ebrtham\u00ebn Linux. K\u00ebt\u00eb her\u00eb, ai u p\u00ebrqendrua n\u00eb planifikuesit e hyrjes\/dalitjes BFQ dhe mq-deadline, t\u00eb cil\u00ebt u treguan si ngushtica n\u00eb t\u00eb pakt\u00ebn rastin e disqeve NVMe. <\/p>\n
Si\u00e7 tregoi studimi i situat\u00ebs, nj\u00eb nga arsyet kryesore p\u00ebr performanc\u00ebn e paoptimizuar t\u00eb n\u00ebnsistemeve t\u00eb planifikuesve t\u00eb input\/output ishte problemet me bllokimet konkurruese (\"lock contention\", p\u00ebrpjekja p\u00ebr t\u00eb marr\u00eb nj\u00eb bllokim q\u00eb mbahet nga nj\u00eb rrjedh tjet\u00ebr). Fal\u00eb masave t\u00eb orientuara p\u00ebr t\u00eb ulur konkurrenc\u00ebn n\u00eb p\u00ebrpunimin e bllokimeve, si serializimi i shp\u00ebrndarjes dhe futja e k\u00ebrkesave, shpejt\u00ebsia e pun\u00ebs s\u00eb planifikuesve u rrit ndjesh\u00ebm n\u00eb disa situata (n\u00eb IOPS). <\/p>\n
N\u00eb testimin e planifikuesit BFQ me mjetin fio, performanca u rrit nga 567K n\u00eb 1551K operacione hyrje\/daljeje n\u00eb sekond\u00eb, nd\u00ebrsa garat p\u00ebr bllokim u zvog\u00ebn nga 96% n\u00eb 30%. N\u00eb rastin e mq-deadline, performanca pas aplikimit t\u00eb patch-ev t\u00eb propozuara duke p\u00ebrdorur nj\u00eb disk NVMe u rrit nga 1070K n\u00eb 2560K operacione hyrje\/daljeje n\u00eb sekond\u00eb (IOPS), nd\u00ebrsa garat p\u00ebr bllokim u zvog\u00ebn nga 94% n\u00eb 23%. \n Burimi: opennet.ru<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"