Subject: Re: Nfs clients get frozen when NFS server crashes...
To: Manuel Bouyer <bouyer@antioche.lip6.fr>
From: Sam <sam@epita.fr>
List: netbsd-users
Date: 10/03/2000 23:40:24
On Tue, Oct 03, 2000 at 09:17:44PM +0200, Manuel Bouyer wrote:
> Date: Tue, 3 Oct 2000 21:17:44 +0200
> From: Manuel Bouyer <bouyer@antioche.lip6.fr>
> To: Sam <sam@epita.fr>
> Cc: netbsd-users@netbsd.org
> Subject: Re: Nfs clients get frozen when NFS server crashes...
> X-Mailer: Mutt 1.0us
> 
> On Tue, Oct 03, 2000 at 06:17:53PM +0000, Sam wrote:
> > You think NetBSD 1.5 is stable enough to serve 500 Pc 24h/24/h ? 
> > Because if it is, we will switch soon. :)
> 
> I don't have that much client so I can't tell. But I have 2 NetBSD NFS
> servers at different sites; one is the alpha, the other a PIII/600 with
> 3 100Mb ethernet. I'm not sure how much machines it serves exactly, should
> be about 40 (mostly linux or NT).

We are setting up a NetbSD1.5_alpha2  Server to make some tests...

> 
> I would suggest setting up a test machine, mount it from you 500 boxes and
> run some tests (cp -R, rm, tar cvf /dev/null, etc) in parallel. This is what
> I did for the alpha and it didn't show up major problems, other than
> the crash of 4 linux clients :)

These are tests that we already done during whole nights :
20 netbsd clients on different 100 Mb networks, taring and copying off and from the 
nfs servers...
It didnt crash the Nfsd....

So we think that it has someting to do with the number of simultaneous access from the clients and/or
a special combination of things causing the nfsd to hang...

> 
> What network do you have (speed and topology) ?

We got 450 Pcs connected to 100 Mb cisco switches (3500/2940) and a catalyst 6000 at the top.
They all mount 5 FreeBsd servers. The two most accessed servers are the ones that have theirs 
nfsd Stopped. It happens most of the time during heavy load, but it already happend fewer 
times under very light load.

> One issue I've seen with NFS and lots of clients is that, under load,
> clients can timeout when the server still have the request queued.
> 
> --
> Manuel Bouyer <bouyer@antioche.eu.org>
> --

I let you guys kmow if i have any news (tests of 1.5Alphas ... )...

Sam

-- 

----
sam@epita.fr
Admin - Sys 
EPITA (Ecole pour l'informatique et les techniques avancees)
Tel: 01 44 08 01 96