Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 5 May 1997 16:38:27 +0400
From:      "Mikhail A. Sokolov" <mishania@demos.su>
To:        "Mikhail A. Sokolov" <mishania@demos.su>
Cc:        hackers@FreeBSD.ORG
Subject:   Re: strange 2.2.1 behaviour.
Message-ID:  <19970505163827.06802@skraldespand.demos.su>
References:  <3.0.32.19970504212053.006a6ee8@mail.hexanet.fr> <19970505000602.51173@skraldespand.demos.su>

next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, May 05, 1997 at 12:06:04AM +0400, Mikhail A. Sokolov wrote:
> On Sun, May 04, 1997 at 09:20:53PM +0200, Christophe Prevotaux wrote:
> > >On Sun, May 04, 1997 at 01:37:01AM +0400, Mikhail A. Sokolov wrote:
> Uhm, the problem now seems to be in kernel level network support, - when we
> turn off nfs's and leave machines alone, they work ok for hours. 
> Like, we achieved load averages ~280-310 with 700 processes running 
> (my favourite find -exec ls -laRt/du -sk. ;-)), - they ate it ok. 

Fun fun fun. We found that there's some problem in the following: 
we have a several sparc server 1000 and ultra 1's (both solaris 2.5.1) in the 
same /32 network which both have numerous interfaces as well.
What correlates, is that when BSD machines are loaded with networking stuff,
like nfs/slirp/ftp/http traffic themselves, they can't survive loads of who 
traffic (513 port), which is issued by  _each_ interface sun's have, (~300/500)
for some reasons, broadly casting, several times (like, 100). Dunno why would
 sun's issue such load of useless traffic, that's what happens. Thanks to 
tcpdump, we see that our (BSD)  machines crash time exactly correlates with
the moments, when network gets 1,5M of who requests in ~5000 packets /sec,  -
they eat those who's ok, when they are not busy themselves, managing _their_
traffic (like in NFS example, they crashed alot more often than non-nfs ones). 
Strange, but we have this port, 513, disabled anyhow.

What we see, I guess, is some kind of buffer overflow, right? If yes, what
paramerts do we increase (on kernel level?)?
 
Thanks, 



-mishania



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19970505163827.06802>