{"id":144005,"date":"2025-09-23T23:11:59","date_gmt":"2025-09-23T21:11:59","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/kompaniya-bytedance-predstavila-svoyo-reshenie-dlya-vypolneniya-neskolkih-yader-linux"},"modified":"2025-09-23T23:11:59","modified_gmt":"2025-09-23T21:11:59","slug":"kompaniya-bytedance-predstavila-svoyo-reshenie-dlya-vypolneniya-neskolkih-yader-linux","status":"publish","type":"post","link":"https:\/\/prohoster.info\/cs\/blog\/news\/kompaniya-bytedance-predstavila-svoyo-reshenie-dlya-vypolneniya-neskolkih-yader-linux","title":{"rendered":"Spole\u010dnost ByteDance p\u0159edstavila sv\u00e9 \u0159e\u0161en\u00ed pro spu\u0161t\u011bn\u00ed v\u00edce jader Linuxu","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
\u010c\u00ednsk\u00e1 spole\u010dnost ByteDance, kter\u00e1 vyv\u00edj\u00ed slu\u017ebu TikTok, zve\u0159ejnila na seznamu pro v\u00fdvoj\u00e1\u0159e j\u00e1dra Linux opravy realizuj\u00edc\u00ed syst\u00e9m Parker (PARtitioned KERnel), kter\u00fd umo\u017e\u0148uje spou\u0161t\u011bn\u00ed n\u011bkolika jader Linux na jednom po\u010d\u00edta\u010di bez pou\u017eit\u00ed tradi\u010dn\u00ed virtualizace. Co se t\u00fd\u010de \u0159e\u0161en\u00fdch \u00faloh, Parker p\u0159ipom\u00edn\u00e1 p\u0159ed n\u011bkolika dny p\u0159edstaven\u00fd projekt Multikernel, ale li\u0161\u00ed se od n\u011bj architekturou a realizac\u00ed. <\/p>\n
V Parkeru se prov\u00e1d\u00ed sektorov\u00e1n\u00ed existuj\u00edc\u00edch jader CPU, pam\u011bti a za\u0159\u00edzen\u00ed. Ka\u017ed\u00fd spou\u0161t\u011bn\u00fd exemplar j\u00e1dra vych\u00e1z\u00ed z jednoho spole\u010dn\u00e9ho obrazu j\u00e1dra, ale m\u016f\u017ee pou\u017e\u00edvat r\u016fzn\u00e9 nastaven\u00ed a optimalizace. Prvn\u00ed na\u010dten\u00e9 j\u00e1dro (Boot Kernel) prov\u00e1d\u00ed distribuci a sektorov\u00e1n\u00ed hardwarov\u00fdch za\u0159\u00edzen\u00ed. Ostatn\u00ed instance j\u00e1dra Linux (Application Kernel) pracuj\u00ed pouze s p\u0159id\u011blen\u00fdmi j\u00e1dry CPU, pam\u011bt\u00ed a za\u0159\u00edzen\u00edmi pro vstup\/v\u00fdstup. <\/p>\n
V po\u010d\u00e1te\u010dn\u00edm variantu pro sn\u00ed\u017een\u00ed re\u017eijn\u00edch n\u00e1klad\u016f a zjednodu\u0161en\u00ed realizace chyb\u00ed \u0159\u00edd\u00edc\u00ed mechanismus a v\u0161echna j\u00e1dra pou\u017e\u00edvaj\u00ed jednu oblast izolace (domain), tj. chyba v jak\u00e9koli instanci j\u00e1dra m\u016f\u017ee v\u00e9st k probl\u00e9m\u016fm v cel\u00e9m syst\u00e9mu. V budoucnu se pl\u00e1nuje zapojen\u00ed samostatn\u00fdch hardwarov\u00fdch mo\u017enost\u00ed pro izolaci instanc\u00ed j\u00e1dra. Jako oblast pou\u017eit\u00ed Parkeru se zmi\u0148uj\u00ed po\u010d\u00edta\u010de s velk\u00fdm po\u010dtem jader CPU, kde p\u0159i \u0161k\u00e1lov\u00e1n\u00ed \u00faloh vznikaj\u00ed probl\u00e9my. <\/p>\n
Pro spr\u00e1vu v Parkeru byl realizov\u00e1n rozhran\u00ed na b\u00e1zi kernfs, a pro na\u010d\u00edt\u00e1n\u00ed dal\u0161\u00edch instanc\u00ed j\u00e1dra do vyhrazen\u00fdch oblast\u00ed pam\u011bti se pou\u017e\u00edv\u00e1 kexec. P\u0159ed spu\u0161t\u011bn\u00edm dal\u0161\u00edch jader prim\u00e1rn\u00ed j\u00e1dro uvol\u0148uje j\u00e1dra CPU, rezervuje fyzickou pam\u011b\u0165 a odpojuje PCI za\u0159\u00edzen\u00ed pro jejich pou\u017eit\u00ed v sektorovan\u00e9m j\u00e1d\u0159e. Pro vstup\/v\u00fdstup by m\u011blo b\u00fdt ka\u017ed\u00e9mu j\u00e1dru p\u0159i\u0159azeno sv\u00e9 PCIe za\u0159\u00edzen\u00ed, jako nap\u0159\u00edklad NVMe \u00falo\u017ei\u0161t\u011b nebo s\u00ed\u0165ov\u00fd adapt\u00e9r. Po spu\u0161t\u011bn\u00ed instance j\u00e1dra nem\u016f\u017ee interagovat s jin\u00fdmi j\u00e1dry. \n Zdroj: opennet.ru<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"