Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 3 Jun 2006 22:35:36 -0700
From:      David Wolfskill <david@catwhisker.org>
To:        current@freebsd.org
Subject:   Re: md /tmp and async mounts
Message-ID:  <20060604053536.GH32476@bunrab.catwhisker.org>
In-Reply-To: <20060604050316.GE61942@gothmog.pc>
References:  <20060523194106.GA46634@xor.obsecurity.org> <20060524203645.GB13500@gothmog.pc> <20060524203747.GA88742@xor.obsecurity.org> <20060524204617.GA13701@gothmog.pc> <20060601002024.GA1453@gothmog.pc> <20060601210655.GA36389@xor.obsecurity.org> <20060601213527.GA53422@gothmog.pc> <20060602215005.GA43170@nowhere> <20060602220724.GA71883@xor.obsecurity.org> <20060604050316.GE61942@gothmog.pc>

next in thread | previous in thread | raw e-mail | index | archive | help

--ZPDwMsyfds7q4mrK
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Sun, Jun 04, 2006 at 08:03:16AM +0300, Giorgos Keramidas wrote:
> ...
> >> I think if you don't have any swap configured, a swap-backed md
> >> will be no worse off than a memory-backed one would.

I'd be rather strongly incline dto agree with that, FWIW.

> > Yeah, it's kind of a poorly chosen name.
>=20
> Should we still revert the default from using -M for tmpmfs=3D"YES" and
> varmfs=3D"YES" in rc.conf?

Ever since /usr/src/etc/rc.d/tmp rev. 1.35 (creation of $tmpmfs_flags),
I have used the specification to avoid -M.  I'd recommend that the
default ought not be -M.

Since Kris posed his query re: -o async, I changed my $tmpmfs_flags to
include it -- both for 6-STABLE and for -CURRENT -- on my laptop, where
I track each of those branches every day that there's a change to the
corresponding source tree.  (Well, save for last weekend, when I was
off-Net for a couple of days....)

While I haven't seen a noticable performance improvement from -o async,
I've certainly not seen any negative impact at all.

FWIW, the one other tweak to $tmpmfs_flags I find useful is to adjust
the inode density to reflect a larger number of (smaller) files in the
/tmp file system.  On the laptop, where the daily buildworlds (and
occasional builds of mozilla...) tend to be the most strenuous workout
it gets, -i4096 seems to be OK.

For CVS servers (which the laptop can be at times, though I prefer to
use more dedicated resource to that sort of task), I find -i1024 to be
preferable.  (Recall the behavior of a CVS server when as "cvs update"
is requested involves building an isomorphic directory hierarchy in
/tmp.  I was pleased to see that a CVS server I had set up had no
problems doing 9 simultaneous "cvs update" runs against /usr/ports.
Absent the -i1024 setting, that would likely have been
less-than-satisfactory.)

Peace,
david
--=20
David H. Wolfskill				david@catwhisker.org
Doing business with spammers only encourages them.  Please boycott spammers.

See http://www.catwhisker.org/~david/publickey.gpg for my public key.

--ZPDwMsyfds7q4mrK
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.0 (FreeBSD)

iEYEARECAAYFAkSCcScACgkQmprOCmdXAD2uhwCfUitjyJF9OvyMItGMGkOE1YEF
D1EAn2RpjB6/gmEl+nv22bdL3onol89I
=OKrH
-----END PGP SIGNATURE-----

--ZPDwMsyfds7q4mrK--



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20060604053536.GH32476>