From owner-freebsd-questions  Tue Apr 27 11:27:24 1999
Delivered-To: freebsd-questions@freebsd.org
Received: from oconnor.inktomi.com (oconnor.inktomi.com [209.1.33.67])
	by hub.freebsd.org (Postfix) with ESMTP id 9CD5814D77
	for <questions@freebsd.org>; Tue, 27 Apr 1999 11:27:22 -0700 (PDT)
	(envelope-from bevans@oconnor.inktomi.com)
Received: (from bevans@localhost)
	by oconnor.inktomi.com (8.8.7/8.8.7) id LAA22681
	for questions@freebsd.org; Tue, 27 Apr 1999 11:56:41 -0700
Message-ID: <19990427115641.A22596@oconnor.inktomi.com>
Date: Tue, 27 Apr 1999 11:56:41 -0700
From: Brennan Evans <bevans@oconnor.inktomi.com>
To: questions@freebsd.org
Subject: limit on max number of mbufs?
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
X-Mailer: Mutt 0.93.2
Sender: owner-freebsd-questions@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.ORG

Hi all,

How do I tune up the maximum number of mbufs on FreeBSD 3.1?
I'm running into what seems to be a 16k buffer limit when trying
to do some FTP tests around here.

thank you,

-- 
brennan

-------------------------------------------------------------------------
Brennan Evans, bevans@inktomi.com, Inktomi Corporation, (650) 653-2990

[ . . . ]

On Tue, Apr 27, 1999 at 12:49:25AM -0700, Brennan Evans wrote:
>
> I already tried the recompile.  USENET seems to indicate that
> mbufs is controled by 4*NMBCLUSTERS, so I would expect something
> larger than the 16k max mbufs that I am currently seeing.  Is
> there another way to bump this up?
>
> I had tried setting NMBCLUSTERS huge before sending the previous
> email. (we've always had maxusers set to 512).
>
>   maxusers        512
>   options         "NMBCLUSTERS=(16*1024)" # huge (and experimental)
>
> BTW:  sysctl -a | grep "tcp.*space"
>       net.inet.tcp.sendspace: 32768
>       net.inet.tcp.recvspace: 32768
>
> >From the output below it appears that the mbuf clusters themselves
> are not saturated, but that I've maxed out on the available mbufs,
> possible?
>

[ . . . ]

> > > proxying FTP causes frequent errors returned to the client.
> > >
> > > Change #1:
> > >
> > > The kernel defaults for ephemeral ports lie in the range.
> > >
> > >   net.inet.ip.portrange.first: 1024
> > >   net.inet.ip.portrange.last: 5000
> > >
> > > The logical range should be somewhat larger [Stevens]:
> > >
> > >   net.inet.ip.portrange.first: 1024
> > >   net.inet.ip.portrange.last: 49000
> > >
> > > Change #2:
> > >
> > > Fixing the first problem leads to another tuning problem.
> > > Still, the rate is so high that we burn up the available
> > > mbufs.  At least now the errors are transient (though
> > > often).
> > >
> > >   ts-bsd2# netstat -m
> > >   16426/16736 mbufs in use:
> > >           358 mbufs allocated to data
> > >           16068 mbufs allocated to packet headers
> > >   288/646/8704 mbuf clusters in use (current/peak/max)
> > >   3384 Kbytes allocated to network (77% in use)
> > >   0 requests for memory denied
> > >   0 requests for memory delayed
> > >   0 calls to protocol drain routines

[ . . . ]


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