Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 8 Aug 1999 22:50:59 +0100
From:      David Malone <dwmalone@maths.tcd.ie>
To:        Warner Losh <imp@village.org>
Cc:        "Jordan K. Hubbard" <jkh@zippy.cdrom.com>, tcobb@staff.circle.net, lightningweb@hotmail.com, freebsd-stable@FreeBSD.ORG, greg@lightningweb.com, jeremy@lightningweb.com, keith@lightningweb.com, criter@lightningweb.com
Subject:   Re: continued crashes with 3.1-Stable
Message-ID:  <19990808225059.B54409@walton.maths.tcd.ie>
In-Reply-To: <199908081755.LAA01058@harmony.village.org>; from Warner Losh on Sun, Aug 08, 1999 at 11:55:31AM -0600
References:  <2602.934003898@localhost> <199908081755.LAA01058@harmony.village.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, Aug 08, 1999 at 11:55:31AM -0600, Warner Losh wrote:
> In message <2602.934003898@localhost> "Jordan K. Hubbard" writes:
> : I wonder why this doesn't happen to Yahoo's SMP boxes then.
> : They run -stable and would be screaming blue murder if their
> : webservers were doing the same thing.  So, for that matter,
> : would Hotmail.
> 
> We run -stable on a SMP boxes at work.  We've seen minor annoyances
> (like times for process not being right, etc), but nothing major.  The 
> one bad bug we've found with NFS happens on UP machines too..

We've found one bug in -stable which we sent Matt a fix for -
unfortunately it uncovers another bug, which we sent Matt a workaround
for. We've been testing the patches for several months and they
have improved the stability of our 3.2 NFS clients alot. The bug
is to do with concurrent writes to files and is easier to provoke
with a SMP machine.

Basically, if you write a program which does lots of short unbuffered
writes, and then do

a.out > /nfs/file&
a.out > /nfs/file&
a.out > /nfs/file&
a.out > /nfs/file&
a.out > /nfs/file&
.
.
.

a few times you'll hang the machines. If you try it on -current you'll
just panic the machine. You'll also see that it takes *ages* to finish
doing these - which is because of the way our buffer locking works.

We also have patches which makes UDP based NFS reply with the correct
source address, which we're cleaning up.

	David.


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




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19990808225059.B54409>