[OpenIndiana-discuss] update on zfs_nocacheflush

Gabriele Bulfon gbulfon at sonicle.com
Mon Oct 10 16:16:10 UTC 2011


I did some more checks.
I used my adaptec controller administration commands to check for the status.
I have 5 logical volumes each on one physical disk.
I saw I had the volumes with wcache disabled, but physical disks with cache enabled.
So, i used the arcconf command to change the physiscal disks and disable cache.
Both each logical volume and each physical disk has Write Cache Disabled (write-through).
System was still with zfs_nocacheflush=1, but because disks has no cache at all,
I exepcted a slow down, but the machine appeared fast as with cash and zfs_nocacheflush.
So I turned nocacheflush=0 again, and I expected no change again, but then I have a slow down....
I'm totally confused......
----------------------------------------------------------------------------------
Da: Gabriele Bulfon
A: Discussion list for OpenIndiana
Data: 10 ottobre 2011 12.07.08 CEST
Oggetto: Re: [OpenIndiana-discuss] use zfs_nocacheflush
Thanx.
Yes, I understand the danger, and I have to say I first tried using NFS async, but that made no
difference to me.
I'm still a bit confused about this nocacheflush flag, though.
Does it refers ONLY to controllers/disks cache or to the zfs system cache too?
I disabled write cache on every disk in the zpool, through the controller bios.
So, I assume there should be no difference between the two flag statuses: I believe the controller
should always return "ok, done" once written, because it has no cache at all. Am I wrong?
Probably I'm wrong, because the performace increased a lot with nocacheflush=1, so maybe there
is other cache involved?
Thanx.
----------------------------------------------------------------------------------
Da: Albert Lee
A: Discussion list for OpenIndiana
Data: 8 ottobre 2011 0.36.02 CEST
Oggetto: Re: [OpenIndiana-discuss] use zfs_nocacheflush
On Thu, Oct 6, 2011 at 5:45 AM, Gabriele Bulfon
wrote:
Hi,
NFS on zfs can be quite a pain with large number of small files.
After playing around it, I discovered this zfs_nocacheflush flag bringing me back to high
performances on NFS.
Questions:
- How much unsafe is this?
This is completely unsafe and can easily lead to data loss unless you
have a battery-backed cache. This setting determines whether the cache
is flushed to complete transactions. NFS operations are synchronous.
Use a log device if you want fast synchronous performance, or a mount
option on the client to make operations async (if you don't mind
losing recent NFS writes in the event of a crash). Disabling atimes
may also help.
-Albert
_______________________________________________
OpenIndiana-discuss mailing list
OpenIndiana-discuss at openindiana.org
http://openindiana.org/mailman/listinfo/openindiana-discuss
_______________________________________________
OpenIndiana-discuss mailing list
OpenIndiana-discuss at openindiana.org
http://openindiana.org/mailman/listinfo/openindiana-discuss


More information about the OpenIndiana-discuss mailing list