From owner-freebsd-hackers Sun Mar 10 20:27:44 2002 Delivered-To: freebsd-hackers@freebsd.org Received: from imr1.aus.deuba.com (bagheera.aus.deuba.com [203.0.62.7]) by hub.freebsd.org (Postfix) with ESMTP id E028737B405 for ; Sun, 10 Mar 2002 20:27:40 -0800 (PST) Received: from imr1.aus.deuba.com by imr1.aus.deuba.com id g2B4RdHR011444; Mon, 11 Mar 2002 15:27:39 +1100 (EST) Received: from merton.aus.deuba.com by imr1.aus.deuba.com id g2B4Rdn2011441; Mon, 11 Mar 2002 15:27:39 +1100 (EST) Received: (qmail 47781 invoked by uid 107); 11 Mar 2002 04:27:39 -0000 Message-ID: <20020311042739.47780.qmail@merton.aus.deuba.com> From: callum.gibson@db.com Subject: Bad network performance after 4.5 upgrade To: hackers@freebsd.org Date: Mon, 11 Mar 2002 10:28:59 +1100 (EST) X-Mailer: ELM [version 2.5 PL2] Sender: owner-freebsd-hackers@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.ORG Hi, I have a FreeBSD box connected to another via a dual-channel isdn router (64+64k). The way this works is that if sustained throughput is detected the second channel on the router (cisco 770) automatically comes up giving throughput of around 14k/sec (or 7k/sec for a single channel). One was running 4.3, the other 4.4. Recently I upgraded the 4.3 machine (which is plugged directly into the router) to 4.5 and noticed an immediate problem with the network upload throughput. Since I was intending to upgrade the other machine anyway I suspended judgement until then. Having done so the problem persists, which is this: as soon as the second channel cuts in the network throughput splutters to a halt. For example, when using scp or ftp it tells me the connection has stalled. The second channel then automatically drops out again due to no network activity (as indicated by the leds). New network sessions can usually be started ok, but it takes quite a while for the scp transfer to recover and the cycle repeats. Initially I suspected the openssh software (since this would have changed with the upgrade), but I have upgraded it and then also noticed the same behaviour with X11 connections, ftp, etc. Additionally, scp transfers between the FreeBSD box and a Solaris 2.6 box across the isdn line behave as expected with the second channel clicking in and staying up, instead of just dying. I've discounted the idea that it's the network driver per se, since it works fine to Solaris and is okay when downloading. Help! Is there some sort of sysctl tweak? I wonder if some sort of timing is being thrown out by the brief delay as the second channel comes up. I witnessed a bit of talk in hackers around December about tcp changes and a bit of recent activity mentioning patches. At this stage, though, it seems like the 4.5 upgrade has stuffed my network throughput in one direction. Here is the ifconfig and dmesg output for the affected end. The isdn router is a cisco 770. xl0: flags=8843 mtu 1500 options=3 inet6 fe80::2c0:4fff:fe60:ad0d%xl0 prefixlen 64 scopeid 0x1 inet 10.179.46.21 netmask 0xfffffff8 broadcast 10.179.46.23 ether 00:c0:4f:60:ad:0d media: Ethernet 10baseT/UTP status: active xl0: <3Com 3c905B-TX Fast Etherlink XL> port 0xec00-0xec7f mem 0xfe008000-0xfe00 807f irq 10 at device 17.0 on pci0 xl0: Ethernet address: 00:c0:4f:60:ad:0d regards, Callum (c)2002 Callum Gibson callum.gibson@db.com Global Markets IT, Deutsche Bank, Australia 61 2 9258 1620 ### The opinions in this message are mine and not Deutsche's ### To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message