You are not logged in.
Pages: 1
I've been using ZFS on a couple of my Arch machines for a few weeks. I had initially used `zfs-dkms` on both machines, both using -ck kernels (one sandybridge, one haswell). However, on my desktop machine I started having horribly slow transfer on my ZFS drives, and eventually tracked the issue down to the wrong scheduler being used on the drive. Namely [bfq] rather than [noop]. Switching to [noop] increased transfer rate by somewhere between ten to fifeteen fold!
I thought it was my use of a non-standard kernel which caused the ZFS drives to be incorreclty using [bfq] rather than [noop], so I switched to the mainline Arch kernel. But I had the same issue, except [cfq] was being set everywhere rather than [bfq].
So then I thought that perhaps `zfs-dkms` was the issue, so I installed `zfs-linux-git` from the archzfs repo (and uninstalled `zfs-dkms`). But I had the same issues w.r.t. [noop] not being set on zfs-only drives.
Now, of course, I can manually set [noop] on the zfs drives no matter which kernel I use (though I'm still trying to figure out a good way of setting it automatically but only for the zfs drives....), but --- as far as I know --- ZFS-only drives should automatically be set as [noop] on ZoL, so I'm wondering if I'm somehow using the 'wrong' approach to ZFS on Arch: I'm worried that the [noop] issue may not be the only issue if I'm somehow just doing things the wrong way w.r.t. ZFS (though there are no other obvious issues).
Last edited by emacsomancer (2017-03-19 02:10:30)
Offline
maybe some help https://github.com/zfsonlinux/zfs/issues/169
Offline
Pages: 1