Changing the default for ZFS atime to off?

[Jeremy Chadwick]

On Sun, Jun 09, 2013 at 05:39:42PM +0100, Steven Hartland wrote:
> 
> ----- Original Message ----- From: "Damien Fleuriot" <ml at my.gd>
> To: "Steven Hartland" <smh at freebsd.org>
> Cc: <fs at freebsd.org>
> Sent: Sunday, June 09, 2013 11:39 AM
> Subject: Re: Changing the default for ZFS atime to off?
> 
> 
> >
> >On 8 Jun 2013, at 20:54, "Steven Hartland" <smh at freebsd.org> wrote:
> >
> >>One of the first changes we make here when installing machines
> >>here to changing atime=off on all ZFS pool roots.
> >>
> >>I know there are a few apps which can rely on atime updates
> >>such as qmail and possibly postfix, but those seem like special
> >>cases for which admins should enable atime instead of the other
> >>way round.
> >>
> >>This is going to of particular interest for flash based storage
> >>which should avoid unnessacary writes to reduce wear, but it will
> >>also help improve performance in general.
> >>
> >>So what do people think is it worth considering changing the
> >>default from atime=on to atime=off moving forward?
> >>
> >>If so what about UFS, same change?
> >
> >I strongly oppose the change for reasons already raised by many
> >people regarding the mbox file.
> >
> >Besides, if atime should default to off on 2 filesystems and on
> >on all others, that would definitely create confusion.
> 
> A very valid point.
> 
> >Last, I believe it should be the admin's decision to turn atime
> >off, just like it is his decision to turn compression on.
> 
> Trying to play devils advocate here; compression is off by default
> because it uses resources and doesn't give a benefit for all cases.

Not to mention ZFS on FreeBSD, specifically WRT compression and dedup,
still lack a separate priority class for their threads.  Info on that:

http://lists.freebsd.org/pipermail/freebsd-fs/2011-October/012718.html
http://lists.freebsd.org/pipermail/freebsd-fs/2011-October/012726.html

While the discussion about atime default is fine/good to have, there are
bigger/more impacting than atime.  (Compression and dedup are something
people *really* want to use, and I understand -- hell, I'd be using
compression if it weren't for the above problem.  It's the sole blocker
for me -- really).

> Is that not the same as atime, and it should be an admins decision
> to turn it on where it's wanted?

While I understand you're playing devil's advocate, you will find I, as
well as most BSD people (in my experience), tend to err on the side of
caution.  That means atime=on as a default.

> >Don't mistake me, we turn atime=off on every box, every
> >filesystem, even on Mac's HFS.
> >Yet I believe defaulting it to off is a mistake.
> 
> That's what prompted me to start this discussion. If a large portion
> of users either disable atime already or would disable atime if they
> knew about it, does that bring into question the current default?

You've just encountered 1 user who sets atime=off on every box they
admin/maintain.

And you have me -- who has atime=on on every box he admins/maintains.

If you're looking for a vote, you won't get one that satisfies everyone,
nor the majority of FreeBSD users -- because most users are not
subscribed to the mailing list, do not visit the forums, etc..  They
install the OS + use it and live happily in their hobbit hole.

I also have no idea how this would impact the commercial companies who
rely on FreeBSD for their enterprise products.  I imagine their feedback
would (should?  Matter of opinion) hold more weight.

> Potentially a better solution would be to make atime an option
> in the installer, as that helps educate admins that the option
> exists, which is potentially the biggest issue here?

As I've stated in some other threads (probably on -stable), I'm all for
people adding options/checkboxes/etc. to bsdinstall to allow more
granularity during installation (vs. having to do things after-the-fact
or the "final shell" prior to rebooting).  If someone wants to add an
atime checkbox (checked == atime enabled) to the filesystem creation
phase, that's fantastic.

But I strongly feel that checkbox needs to default to checked/enabled,
solely so there are no "unwanted surprises" since we have no idea what
software they'll be using on the system.

There is also (still) the concern of POSIX compliance, which the BSDs
have historically been very strict about.  I guess you can hash that out
with Bruce.

Honestly the relatime thing from Linux sounds like a decent compromise.

-- 
| Jeremy Chadwick                                   jdc at koitsu.org |
| UNIX Systems Administrator                http://jdc.koitsu.org/ |
| Making life hard for others since 1977.             PGP 4BD6C0CB |