From owner-freebsd-net Tue Nov 28 5:30:52 2000 Delivered-To: freebsd-net@freebsd.org Received: from iguana.aciri.org (iguana.aciri.org [192.150.187.36]) by hub.freebsd.org (Postfix) with ESMTP id B916E37B400 for ; Tue, 28 Nov 2000 05:30:49 -0800 (PST) Received: (from rizzo@localhost) by iguana.aciri.org (8.11.1/8.11.1) id eASDUO247786; Tue, 28 Nov 2000 05:30:24 -0800 (PST) (envelope-from rizzo) From: Luigi Rizzo Message-Id: <200011281330.eASDUO247786@iguana.aciri.org> Subject: Re: Strange lockups with Dummynet In-Reply-To: from Ferdinand Goldmann at "Nov 27, 2000 10:40: 3 pm" To: ferdl@atommuell.oeh.uni-linz.ac.at (Ferdinand Goldmann) Date: Tue, 28 Nov 2000 05:30:24 -0800 (PST) Cc: freebsd-net@FreeBSD.ORG X-Mailer: ELM [version 2.4ME+ PL43 (25)] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-freebsd-net@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org What OS version are you using, what ipfw/dummynet configuration, how much memory is on the system. cheers luigi > > I am experiencing strange crashes on a machine which is being heavily used > as a traffic shaper serving about 300 clients. The machine itsself is just > an old Pentium machine, NICs are: > > tx0: port 0x6200-0x62ff mem 0xe1000000-0xe1000fff irq 11 at device 18.0 on pci0 > miibus0: on tx0 > nsphy0: on miibus0 > nsphy0: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, auto > tx0: address 00:e0:29:39:bb:ab, type SMC9432BTX > xl0: <3Com 3c905C-TX Fast Etherlink XL> port 0x6300-0x637f mem 0xe1001000-0xe100107f irq 9 at device 19.0 on pci0 > xl0: Ethernet address: 00:50:da:0d:a8:a7 > miibus1: on xl0 > xlphy0: <3c905C 10/100 internal PHY> on miibus1 > xlphy0: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, auto > > Occasionally, I see these in the logs: (I hear this is due to hardware being > too slow to respond to feed the NIC, which could be possible (P166)) > > xl0: transmission error: 90 > xl0: tx underrun, increasing tx start threshold to 120 bytes > > I have set the following kernel parameters: > maxusers 96 > options NMBCLUSTERS=4096 > > I beefed up NMBCLUSTERS because FreeBSD ran out of them with the default > setting. > > Currently: > # netstat -m > 637/1472/16384 mbufs in use (current/peak/max): > 187 mbufs allocated to data > 450 mbufs allocated to packet headers > 186/934/4096 mbuf clusters in use (current/peak/max) > 2236 Kbytes allocated to network (23% in use) > > I have configured between 3 and 4 pipes on this machine and declared > a small ruleset which should shape the traffic according to my > expectations. > > Today, I experienced the following: Upon adding a new pipe definition, > the machine locked maybe an eyeblink after the command was set off. > I.e., the machine was not ping'able anymore, and the console was > dead. However, the activity LEDs of the NICs were still flashing. > No kernel entries, nothing... > This has happened already once after rising the NMBCLUSTERS value. > Being paranoid, I have been logging this value to a file, but it does not > seem to rise significantly before the crash. > > [Q]: Can the change of NMBCLUSTERS be the cause for my lockup > problem, or could this be a dummynet problem? How do people running > sites with lots of traffic setting the values of maxusers and > NMBCLUSTERS to be safe, and what would be the maximum value for > maxusers (I heard long it is instable > 128)? > General question, how stable is dummynet under heavy load, any experiences > there? > BTW, the machine is also doing port forwarding of HTTP connects to a proxy. > > TIA for any hints, > Ferdinand Goldmann > > > > To Unsubscribe: send mail to majordomo@FreeBSD.org > with "unsubscribe freebsd-net" in the body of the message > To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-net" in the body of the message