Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 21 Jul 2014 14:44:48 -0400
From:      Paul Kraus <paul@kraus-haus.org>
To:        questions@freebsd.org
Subject:   Re: deciding UFS vs ZFS
Message-ID:  <B128508A-8830-4D36-B4AD-E285BDF9D5BC@kraus-haus.org>
In-Reply-To: <20140713190308.GA9678@bewilderbeast.blackhelicopters.org>
References:  <20140713190308.GA9678@bewilderbeast.blackhelicopters.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On Jul 13, 2014, at 15:03, Michael W. Lucas <mwlucas@michaelwlucas.com> =
wrote:

> My virtualization system runs KVM, so I use UFS on VMs. Restoring
> ZFS disk images via dd can be problematic.

I use ZFS for the host and (generally) UFS on the guests. By using one =
ZFS dataset per VM and taking frequent (hourly) snapshots I have an easy =
way to roll a VM back to a stable state.

> For larger boxes running on real iron, I use ZFS.
>=20
> But there's a whole range of conditions between these two. And the way
> to fill in the gray spaces is to ask.
>=20
> So, how do you decide to use which filesystem?

I think this decision tree breaks in a number of ways :-)=20

Desktop vs. Server is the first=85

In my world physical servers get ZFS for all of the various reasons =
others have points out, but the biggest is the ensured data integrity. =
While not perfect (nothing is), it is much closer than any other open =
source solution at this point.

I generally do not run FreeBSD as a Desktop (nor OmniOS or SmartOS), so =
I don=92t have a strong answer for the Desktop case.

Once you go down the Server branch there are more branches=85

Physical vs. Virtual

I already stated that I like ZFS on physical hardware. ZFS gives me many =
handles to tune (almost too many), between:

basic vdev layout and configuration: mirrors, raidz<n>, how many vdevs, =
etc.

compression: I have seen compression make a 5 fold increase in =
performance, I have also seen it reduce performance, it all depends on =
your work load, CPU horsepower, and memory bandwidth.

dedupe: for certain very specific workloads it can make a huge =
improvement, for all the others it generally causes more trouble than it =
is worth

default block size: this one generates more debate than all of the =
others combined, suffice it wo say, test with *your* data and choose =
wisely

L2ARC and ZIL devices: lots more misinformation out there; mirroring =
L2ARC is only occasionally of benefit, ZIL *must* be mirrored, but only =
benefits SYNC writes. Know your workload and adjust accordingly.

For Virtual systems I have used both (ZFS and UFS) and have had no real =
negative (or positive) experiences.

Then you get into the Type of Server (be it physical or virtual):

End User Mail servers (IMAP, etc.) represent a very different workload =
than traditional DB servers, but I still like ZFS for the ease of =
management and tune-ability.

NOTE: I spent over 10 years managing Solaris systems and storage and ZFS =
was a welcome change that greatly simplified storage management =85 at a =
cost. I had many more ways to recover a SLVM/UFS filesystem than I do a =
ZFS dataset. Part of that is maturity, when ZFS is as old as UFS I=92m =
sure it=92ll be there too :-=3D)

--
Paul Kraus
paul@kraus-haus.org




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?B128508A-8830-4D36-B4AD-E285BDF9D5BC>