Re: Strange things on GBit / 1000->100 / net.inet.tcp.sack

From: Jon Noack (noackjr_at_alumni.rice.edu)
Date: 09/27/04

  • Next message: gkullak_at_fi.uba.ar: "ipnat of ipfilter crash with too many mapping?" 
    Date: Mon, 27 Sep 2004 09:21:59 -0500 (CDT)
To: "Raphael H. Becker" <rabe@p-i-n.com>

    Raphael H. Becker wrote:
    > Hi Jon,
    >
    > On Sat, Sep 18, 2004 at 05:37:10PM -0500, Jon Noack wrote:
    >> Have you tried disabling SACK (net.inet.tcp.sack.enable) on the 5.3
    >> machine?
    >
    > Tried that now.
    >
    > case 1000->100:
    >
    > sack.enable:1 300kbytes/sec
    > sack.enable:0 1.4MBytes/sec (should be around 10MBytes/sec)
    >
    > say: slightly better performance without sack.

    Interesting. While I don't think that sack is to blame here, it appears
    to aggravate the situation. Your network setup is not going to
    significantly benefit from sack, so you could leave it off to help...

    > case 1000->1000:
    >
    > sack.enable:1 76.82 MBytes/sec
    > sack.enable:0 76.80 MBytes/sec
    >
    > say: exactly no difference.
    >
    > The FTP-server is a 5.3-BETA4, the client(100) still a 4.10-p2 and
    > client(1000) a 5.3-BETA5.
    >
    > Any idea?
    >
    > I still guess the rubbish switch has a problem with its store&forward
    > engine and cannot buffer much data, maybe broken by design.

    I think others may have asked this, but have you tried using a different
    switch? Also, what happens if you force the card to 100Mbit? Do you
    still get poor performance?

    > On the other hand, freebsd 5.3 seems not to recognize this and I don't
    > know where in the network stack the traffic is limited to effectivly
    > 100MBit/sec, when sending to a 100MBit-node. With 4.10 as GBit-sender
    > the transfer to a 100MBit-Node works perfectly (10.5 MBytes/sec ftp).

    Yeah, ideally we would gracefully handle finicky hardware.

    > The differences on sender's side are:
    >
    > a) 5.3-BETA (poor) 4.10-p2 (perfect)
    > b) <Broadcom BCM5703 Gigabit Ethernet, ASIC rev. 0x1002> (poor) vs
    > <BCM5701 10/100/1000baseTX PHY> (perfect)
    >
    > HTH
    > Raphael Becker
    >
    > Maybe you have an idea about the switch-specs:
    > http://www.netgearinc.co.jp/support/pdf/gs516t_manual.pdf page 22 of 34

    Sorry, I'm not knowledgeable enough to comment intelligently on switch specs.

    Jon

    _______________________________________________
    freebsd-net@freebsd.org mailing list
    http://lists.freebsd.org/mailman/listinfo/freebsd-net
    To unsubscribe, send any mail to "freebsd-net-unsubscribe@freebsd.org"

  • Next message: gkullak_at_fi.uba.ar: "ipnat of ipfilter crash with too many mapping?"