Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 8 Mar 2001 21:47:49 -0800
From:      "Ted Mittelstaedt" <tedm@toybox.placo.com>
To:        "Doug Young" <dougy@gargoyle.apana.org.au>, "Denis J. Cirulis" <monster@okb.lv>, <freebsd-newbies@FreeBSD.ORG>
Subject:   RE: About Unix
Message-ID:  <003d01c0a85c$7955d620$1401a8c0@tedm.placo.com>
In-Reply-To: <024301c0a7de$930db020$847e03cb@apana.org.au>

next in thread | previous in thread | raw e-mail | index | archive | help
>-----Original Message-----
>From: owner-freebsd-newbies@FreeBSD.ORG
>[mailto:owner-freebsd-newbies@FreeBSD.ORG]On Behalf Of Doug Young
>
>There is a technical basis to this stuff ..... apprently linus & his
>followers
>chose to adopt a different way of writing files than with "proper"unixes in
>order to gain a bit more speed, whereas BSD / Solaris / SCO / HP-UX
>etc considered reliability of higher significance than outright speed. This
>issue has been discussed several times in the 'questions" list.
>

The difference is that, by default, Linux filesystems are mounted async,
FreeBSD filesystems are mounted sync.

However, this is just a default.  It is trivial to change on both system,
and if you have Linux systems that are going to be in environments where
they are going to be regularly restarting, then it's a lot easier to
change the config to have the automounter mount the Linux filesystems
synchronously.

In any case, all of this is begging 2 very important questions:

1) Why don't you organize your systems to be resistant to this?

2) Why don't you correct the environment so the systems don't have to
restart.

As far as #1 is concerned, I manage a Usenet news server that is very busy.
About once every 2-3 months it gets a SCSI bus error and reboots itself.
After the first one of these I changed the system so that when it does
reboot itself, that there's not a problem.

You see, the issue with uncontrolled shutdowns is this.  If the partition
(note partition, not filesystem) is quiescent during a uncontrolled
shutdown, when it is fsck'd during reboot, there won't be any corruption -
and fsck will mark it clean and remount it.

This leads to an obvious solution - you arrainge your filesystem mount
points so that anything that is being written is NOT on a filesystem
containing startup scripts, (typically in /etc) or on a partition that's
automounted.

For example, with FreeBSD, the default mount points are to put /etc and / on
the same partition.  Fine - but /tmp is created on /, and /tmp is usually
going to be in use during an uncontrolled shutdown.

What I did with my news server is /var is on it's own partition, and all
logs in it are softlinked to another disk.  /tmp and /usr/tmp are also
softlinked to this disk.  The filesystems on this disk are NOT automounted.

If the system crashes and reboots itself, then /, /var, and /usr are all on
partitions that are NEVER written to during normal operation, thus they are
always quiescent, and they always come back up with no problem.  I can then
Telnet into the system and manually run fsck on the other disks.  Granted,
it's a nuisance because the log and temp directories are unavailable during
this limited maintainence mode, but the system won't deny me access.

Once the rest of the disks are clean, I mount them, then restart syslogd and
the other programs that need to be started and away we go.  No need to be
physically at the system to do all this, nor is sync/async mounting an
issue.

Now, as far as #2 is concerned, with the exception of my news server, none
of my other servers ever have uncontrolled shutdowns.  This is because of
several things.  First, all servers have their own UPS's and are plugged
into the sense port of the UPS, and if the UPS goes onto battery for too
long, the server does a controlled shutdown.  The servers and UPS's are also
all on remote reboot switches.  Secondly, if I find a flaky server I work
with it until I fix it or scrap it.  I tolerate the news server because I
know that the problem is a software driver bug and I have not yet gotten
time to rebuild it and fix the bug.  (news servers typically take a long
time to rebuild and tune)


Ted Mittelstaedt                      tedm@toybox.placo.com
Author of:          The FreeBSD Corporate Networker's Guide
Book website:         http://www.freebsd-corp-net-guide.com


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-newbies" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?003d01c0a85c$7955d620$1401a8c0>