Hi Paul, Regarding the IDLE state waiting: - yes, this could/should be backported to any version which already has swait_event_idle() macro(s) - and so also TASK_IDLE - defined. About our fio driven tests - we used fio "ini" files similar to: [global] description=Random Read/Write Access Pattern bssplit=512/8:1k/4:2k/4:4k/30:8k/20:16k/10:32k/8:64k/12:128k/4 cpus_allowed=0-31 cpus_allowed_policy=split numa_mem_policy=local fadvise_hint=0 rw=randrw direct=1 size=100% random_distribution=zipf:1.2 percentage_random=80 ioengine=libaio iodepth=128 numjobs=1 gtod_reduce=1 group_reporting ramp_time=60 runtime=300 time_based [job0] filename=/dev/disk/by-name/vol0 [job1] filename=/dev/disk/by-name/vol1 ... and so on for >= 100 logical volumes. Best regards, On Tue, Mar 5, 2024 at 3:14 PM Paul Menzel <pmenzel@xxxxxxxxxxxxx> wrote: > > Dear Florian-Ewald, > > > Thank you for your quick reply. > > Am 05.03.24 um 13:25 schrieb Florian-Ewald Müller: > > > Regarding the INTERRUPTIBLE wait: > > - it was introduced that time only for _not_ contributing to load-average. > > - now obsolete, see also explanation in our patch. > > Yes, I understood that. My question was regarding, if this should be > backported to the stable series. > > > Regarding our tests: > > - we used 100 logical volumes (and more) doing full-blown random rd/wr > > IO (fio) to the same md-raid1/raid5 device. > > Do you have a script to create these volumes and running the fio test, > so it’s easy to reproduce? > > Was it a spinning disk or SSD device? > > > Kind regards, > > Paul
