Date: Thu, 14 Jun 2001 09:34:27 +0100 From: Nick Barnes <Nick.Barnes@pobox.com> To: "Robert L Sowders" <rsowders@usgs.gov> Cc: "Robin P. Blanchard" <Robin_Blanchard@gactr.uga.edu>, stable@FreeBSD.ORG Subject: Re: gigabit woes Message-ID: <7804.992507667@thrush.ravenbrook.com> In-Reply-To: Message from "Robert L Sowders" <rsowders@usgs.gov> of "Wed, 13 Jun 2001 18:03:32 PDT." <OF532AACAB.95369B84-ON88256A6B.00050D29@wr.usgs.gov>
next in thread | previous in thread | raw e-mail | index | archive | help
At 2001-06-14 01:03:32+0000, "Robert L Sowders" writes: > Ok now your into something I've seen before. > > Had the same problem here with 100mb cards. ifconfig showed them set at > 100mb full duplex. I thought everything was fine. Nope. It was trying > to auto negotiate with the other nic, which was also trying to auto > negotiate. It ended up with one going to full duplex and the other going > to something else. Hence the slow performance. The problem went away > when we forced one of the cards to 100mb full duplex. > > Sorry don't remember how to do this. But it's been discussed in the list > before, do a search for full duplex and force and you should find it. I also had something like this with a 100Mb card. The card seemed to get in a muddle, so hardly any bits were going through. Rebooting didn't fix it, and it wasn't a kernel problem. I fixed it in the end by forcing the media type with ifconfig (maybe "ifconfig <if> media 100baseTX"?). Nick B > > > > > > "Robin P. Blanchard" <Robin_Blanchard@gactr.uga.edu> > Sent by: owner-freebsd-stable@FreeBSD.ORG > 06/13/2001 10:36 AM > > > To: stable@freebsd.org > cc: > Subject: Re: gigabit woes > > yet more findings: > > entirely removed the intel 100Mb nic as well as the 3com 1000Mb nic. > installed a 3com 100Mb nic (device xl); edited my kernel appropriately, > removed my old kernel build-tree, rebuilt the kernel and rebooted. > same test, still at 2.67 MB/s. both ends are showing no errors whatsoever > and are both reporting connection as 100MB full-duplex. > > this is becoming more and more bizarre. > input is appreciated. > > > > new findings on this problem... > > > > so i decided to try the card(s) (3com and intel) in another, older > > dell poweredge 4300. before installing the card, i wanted to test > > the performance on the already installed 100Mb intel card (fxp). > > using the same test, i was getting around 10.3MB/s. that done, i > > reconfigured the kernel to support the ti device, disabled > > the fxp ifconfig entry in rc.conf, enabling ti instead, shut > > the box down, installed the card and rebooted. i ran the same > > test again (now using the gig interface) and wound up with a whopping > > 2.67 MB/s. everything looks fine at the switch, and freebsd is not > > complaining. so i decided to revert back to the 100Mb interface > > (at least i'll get 10 MB/s) by disabling the ti ifconfig entry in > > rc.conf and enabling fxp again. reboot. run my test again (again on fxp) > > and i find i'm getting 2.67 MB/s now -- consistently. what? > > so i remove the ti device from the kernel config and remove the > > card from the system. still 2.67 MB/s on the fxp interface. > > i am very curious to hear any hypotheses on this. > > > > and yes, i also only get 2.67 MB/s on the gig interface if it is alone > > in the system. > > > > > ok. currently in our dell poweredge 4350 there is a 3com 3c985b with > > > alteon-2 chipset plugged into a 64-bit pci slot. the kernel is built > > > to use the ti driver (with kernel NMBCLUSTERS set to 16896). > > > this is directly connected to a gig port on our extreme black diamond. > > > a basic ip test of ftping to a known functional gig interface > > > (an SGI origin 2000 also directly attached to our black diamond -- the > > > same test on this interface with a similar interface on a separate > > > module of this SGI box yields 12.183MB/s) yields a mere 2.37MB/s. > > > both the gig port on the black diamond and the nic in the freebsd box > > > are auto-negotiating. > > > > > > relevant /etc/sysctl.conf: > > > net.inet.tcp.rfc1323=1 > > > net.inet.tcp.sendspace=65536 > > > net.inet.tcp.recvspace=65536 > > > net.inet.udp.sendspace=65536 > > > net.inet.udp.recvspace=65536 > > > > > > relevant /etc/rc.conf: > > > icmp_drop_redirect="YES" > > > inetd_enable="NO" > > > kern_securelevel_enable="NO" > > > moused_enable="NO" > > > nfs_client_enable="YES" > > > ntpdate_enable="YES" > > > ntpdate_flags="10.10.10.11" > > > sendmail_enable="NO" > > > sshd_enable="NO" > > > tcp_drop_synfin="YES" > > > tcp_extensions="YES" > > > tcp_keepalive="YES" > > > usbd_enable="NO" > > > > > > we get similarly poor performance if we swap the 3com with an intel. > > > > > > more performance tuning suggestions are extremely welcomed. > > > thanks in advance. > > > > -- > > ------------------------------------ > > Robin P. Blanchard > > IT Program Specialist > > Georgia Center for Continuing Ed. > > fon: 706.542.2404 fax: 706.542.6546 > > email: Robin_Blanchard@gactr.uga.edu > > ------------------------------------ > > > > To Unsubscribe: send mail to majordomo@FreeBSD.org > > with "unsubscribe freebsd-stable" in the body of the message > > -- > ------------------------------------ > Robin P. Blanchard > IT Program Specialist > Georgia Center for Continuing Ed. > fon: 706.542.2404 fax: 706.542.6546 > email: Robin_Blanchard@gactr.uga.edu > ------------------------------------ > > To Unsubscribe: send mail to majordomo@FreeBSD.org > with "unsubscribe freebsd-stable" in the body of the message > > > > > To Unsubscribe: send mail to majordomo@FreeBSD.org > with "unsubscribe freebsd-stable" in the body of the message > To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?7804.992507667>