From owner-freebsd-net Thu Mar 25 14: 8:25 1999 Delivered-To: freebsd-net@freebsd.org Received: from hp9000.chc-chimes.com (hp9000.chc-chimes.com [206.67.97.84]) by hub.freebsd.org (Postfix) with ESMTP id 2293E14E37; Thu, 25 Mar 1999 14:08:20 -0800 (PST) (envelope-from billf@chc-chimes.com) Received: from localhost by hp9000.chc-chimes.com with SMTP (1.39.111.2/16.2) id AA121217578; Thu, 25 Mar 1999 16:32:58 -0500 Date: Thu, 25 Mar 1999 16:32:58 -0500 (EST) From: Bill Fumerola To: Jesse Cc: freebsd-net@FreeBSD.ORG, freebsd-hackers@FreeBSD.ORG Subject: Re: 3.1-STABLE dies on 40+ connects In-Reply-To: Message-Id: Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-net@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org [ this really shouldn't be x-posted ] On Thu, 25 Mar 1999, Jesse wrote: > > i trust you -- i suspect there is some piece of the code somewhere > > which does not check for mcopy/mpullup/etc failures. > > > > in fact it would be nice to know if there is some reproducible way to > > trigger these crashes because i think this is a problem that ought to > > be fixed in a better way than overallocating resources. > > Okay, I updated the NMBCLUSTERS to 4096. This allowed me to get 40 clients > on successfully. Here's an netstat -m with 40 clients connected: > > leaf:~# netstat -m > 1076/1792 mbufs in use: > 1025 mbufs allocated to data > 51 mbufs allocated to packet headers > 1023/1620/4096 mbuf clusters in use (current/peak/max) > 3464 Kbytes allocated to network (62% in use) Wow that's high. > > So I tried 60 clients. It crashed after the 56 client connected. I was > doing netstat -m the entire time until the moment it crashed. The last one > showed: > > leaf:~# netstat -m > 4637/4704 mbufs in use: > 4565 mbufs allocated to data > 72 mbufs allocated to packet headers > 4564/4604/4096 mbuf clusters in use (current/peak/max) > 9796 Kbytes allocated to network (99% in use) Crunch. > So it definitely appears to be an mbuf issue. Definitely. Could you get some statistics on mbuf/client, for instance, which we could use to determine if the climb is linear or exponential. > Is it normal for something to use mbuf's so quickly? Each client is being > sent a 128kbps stream. I know sites like ftp.cdrom.com transfer MUCH more > than this per second.. soo.. This is certainly a bug of some sort. > It'd of course be nicer to see FreeBSD be 'fixed' to not crash in this > situation, but is there anything that the authors of icecast can do to > help reduce their mbuf usage? Does the machine panic or just crash, that would at least be the first step is just to get FreeBSD to panic. > Lastly, is it safe to raise my nmbclusters higher than 4096? Like 10000? > What are the downsides? Death or memory problems I can't remember which. - bill fumerola - billf@chc-chimes.com - BF1560 - computer horizons corp - - ph:(800) 252-2421 - bfumerol@computerhorizons.com - billf@FreeBSD.org - To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-net" in the body of the message