Dear all. I have zpool/zvol - as a local storage (with l2arc+cache). Which best IO Scheduler in a common case? Is the CFQ best or not in this situation?
Best IO Scheduler for local zfs storage
- Thread starter VGusev2007
- Start date
Depends on your disks: spindels do require a real scheduler due to physical limitations, SSDs normally don't.
Dear LnxBil, I have classical HDD (with SSD l2arc+cache). Do I need to change scheduler?
What is currently used there? Normally - as I read about this issue - ZFS is setting the scheduler accordingly.
How can I check it for zfs?
For zfs I have:
Code:
root@dve-01:~# cat /sys/block/zdXX/queue/scheduler
noneFor HDD I have:
Code:
root@dve-01:~# cat /sys/block/sdc/queue/scheduler
noop [deadline] cfqBy checking I see, that I have noop everywhere where disks were setup with default install and manual SLOG, L2ARC later. HDD or SSD, does not matter.
Unused disks have deadline. I wonder if it will change once I create a ZFS pool on them automatically to noop.
Unused disks have deadline. I wonder if it will change once I create a ZFS pool on them automatically to noop.
So, if I understand you, I don't need to wore about it. Is it good for zfs with noop and deadline for underline hdd?
I _think_ noop should be used always where there are no other filesystems on the same drive and preferably even when they are.
Wrong. If yours storage have anothers layers betheen the storage layers(who will use an unknown or others disk queue like ssd) then the best bet is noop. But if you use let say rotational disks, is better to use deadline (who have 2 differnst queue for writes 40% and for reads 60%).
Any of them noop/dedline will have 0% impact if you do not saturate the I/O disk capacity. Also many others details can impact on this (cache, iops, latency, etc)
As a side note, you can use in OS let say noop, but for zfs you can use dedline.