Date: Sun, 4 May 1997 14:01:28 +0400 (MSD) From: bag@sinbin.demos.su (Alex G. Bulushev) To: stesin@gu.net Cc: mishania@demos.su, hackers@freebsd.org, isp@freebsd.org Subject: Re: strange 2.2.1 behaviour. Message-ID: <199705041001.OAA03719@sinbin.demos.su> In-Reply-To: <Pine.BSF.3.95q.970504115441.18592G-100000@trifork.gu.net> from Andrew Stesin at "May 4, 97 11:57:12 am"
next in thread | previous in thread | raw e-mail | index | archive | help
>
> Hi,
>
> On Sun, 4 May 1997, Mikhail A. Sokolov wrote:
>
> > Intel EtherExpress Pro 10/100B (two, 100Mb full duplex)
> > nfs client
> ^^^^^^^^^^^ !!!
> > network activity is 200-400in/200-400out 1k packets/sec
> > Intel EtherExpress Pro 10/100B (one, 100Mb full duplex)
> > nfs client
> ^^^^^^^^^^^ !!!
> > network activity is also some kind of 200-400in/200-400out packets/sec
> > crashes every 5-30 minutes.
> > Intel EtherExpress Pro 10/100B (two, 100Mb full duplex)
> > nfs server
> ^^^^^^^^^^ and now -- gmmm...
> > network activity 500-1000in/500-1000out packets/sec
> > crashed once 24-48 hr
>
> As for me, I'd try to avoid the whole NFS stuff and see what'll
> happen.
we test it witout nfs, the same results :(
critical is a number of network packets/sec (not traffic)
when packets/sec > 100-300 box's crashed periodicaly
for some box's (with asus mother board) critical is > 1000 packets/sec
we see this in 2.1, 2.1.5 and now in 2.2.1-RELENG
this is not fully a hardware problem, we test this feature
on several box's (HP PC, DEC PC, home made box with asus) and
have the same results ... difference is in a time without
reboots (from 5 min to week) and critical number of packets/sec
may be this is problem in fbsd network layer or pci networks
card drivers ... (we check if_de and if_fxp)
another problem begins in 2.2, on hightly loaded server
with squid (300 cliens)
ircd (50 clients) simultaniously
ftpd (100 clients)
... etc ...
number of mbuf clusters increases permanently, and as a result,
box crashed with " ... out of mbuf ... " message ...
this bug or feature of fbsd >= 2.2 (mbuf.h changed in 2.2)
now we use NMBCLUSTERS=12288 ( ~1 week to work ... )
we report this problem several weeks ago ...
Alex
>
> Best regards,
> Andrew Stesin
>
> nic-hdl: ST73-RIPE
>
>
>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199705041001.OAA03719>