《Chen yu - Linux scheduler scalability on many core system.pdf》由会员分享,可在线阅读,更多相关《Chen yu - Linux scheduler scalability on many core system.pdf(18页珍藏版)》请在三个皮匠报告上搜索。
1、Linux scheduler scalability on many core systemIntel Software and Advanced Technology GroupAgenda Challenge on many core system Proposals to improve scheduler scalability Call for actionChallenge on many core system Challenge on many core system Task wakeup optimizations SIS_UTIL(merged upstream)SIS
2、_SHORT(under discussion)Idle load balance optimization ILB_UTIL(under discussion)ILB_FAST(under discussion)6%idle balance overhead drop in synthetic benchmarkResource aware task placement Task group load_avg update optimization(merged upstream)SIS_CACHE(under discussion)Cluster base task wakeup(merg
3、ed upstream)+48%throughput hackbench Per above discussion,task migration hurts performance on large server,mainly due to the breakage of cache locality.Introduce SIS_CACHESIS_CACHE+75%will-it-scalecontext_switch1 Cluster based task wakeup+16%netperf throughputOpen question What is the kernel scalabi
4、lity issue in production environment?Seek for collaborate opportunity.Reference commit 70fb5ccf2ebb(sched/fair:Introduce SIS_UTIL to search idle CPU based on sum of util_avg)commit 1528c661c24b(sched/fair:Ratelimit update to tg-load_avg)commit 8881e1639f1f (sched/fair:Scan cluster before scanning LLC in wake-up path)SIS_SHORT:https:/lore.kernel.org/lkml/ you!