Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 8 Jan 2013 19:58:50 -0800
From:      Freddie Cash <fjwcash@gmail.com>
To:        John <jwd@freebsd.org>
Cc:        FreeBSD Filesystems <freebsd-fs@freebsd.org>
Subject:   Re: rc.d script for memory based zfs intent log
Message-ID:  <CAOjFWZ5eC=0TyrD5c9A267V1e6uU4jmssH7C1Pcnku_o=ubtJQ@mail.gmail.com>
In-Reply-To: <20130109031834.GA14386@FreeBSD.org>
References:  <20130109023327.GA1888@FreeBSD.org> <20130109.115240.1198411557684741197.hrs@allbsd.org> <20130109031834.GA14386@FreeBSD.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On Jan 8, 2013 7:18 PM, "John" <jwd@freebsd.org> wrote:
>
> ----- Hiroki Sato's Original Message -----
> > John <jwd@freebsd.org> wrote
> >   in <20130109023327.GA1888@FreeBSD.org>:
> >
> > jw> Hi Folks,
> > jw>
> > jw>    Here's an rc.d script that provides a nice performance boost on
> > jw> ZFS/NFS based file servers. It also helps in other areas not
specific
> > jw> to NFS.
> > jw>
> > jw>    It attaches the log device at system startup and removes it at
> > jw> system shutdown time. Example;
> > jw>
> > jw> memzil_pools="tank"
> > jw> memzil_bootfs="YES"
> > jw> service memzil onestart
> > jw> zpool status tank
> > jw> service memzil onestop
> > jw>
> > jw>    This configuration provides a nice performance boost especially
to
> > jw> NFS, but also helps in other areas not specific to NFS.
> > jw>
> > jw>    Please DO NOT USE this script if your system is not UPS backed,
preferably
> > jw> with dual power supplies on separate circuits. If your system
crashes you
> > jw> may lose data. The script contains information on recovery.
> > jw>
> > jw> http://people.freebsd.org/~jwd/memzil.txt
> > jw>
> > jw>    Comments/Improvements appreciated.
> >
> >  Why is simply setting sync=disabled to the ZFS dataset not enough?
>
> As you refer to, my understanding is that sync=disabled is at the dataset
> layer. The zil approach is at the zpool layer - sync=disabled would be
> nice at the zpool layer.

Set it on the root dataset (meaning the dataset with the same name as the
pool) and every child dataset will pick it up via inheritance.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAOjFWZ5eC=0TyrD5c9A267V1e6uU4jmssH7C1Pcnku_o=ubtJQ>