• Interesting iperf UDP behaviour with new kernel/Linux release

    From freeman.wang@gmail.com@1:0/0 to All on Tue Jul 8 03:57:26 2014
    Hi

    I'm trying to figure out what is causing iperf UDP test to some kind of 'fl=
    ow controlled' with the new Fedora 18/Kernel 3.x.

    I tested with the same iperf v2.0.5, same command on the client side:

    iperf -c _server_ip_addr -u -b 4000m -i 1

    With Fedora 14, the client side shows 4Gbps of packets are pushed to the dr= iver and the server side shows huge packet loss. With Fedora 18, the client=
    side seems automatically adjusts its TX rate to the capacity of the link a=
    nd the server side shows almost 0 packet loss.

    I'm very curious what is changed between Fedora 14 and Fedora 18 to give UD=
    P such a different behaviour, and is it the future of UDP implementation?=
    =20

    Thanks
    FRW

    --- MBSE BBS v1.0.1 (GNU/Linux-i386)
    * Origin: The Kofo System II BBS telnet://fido2.kofobb
  • From Rick Jones@110:110/2002 to All on Tue Jul 8 15:17:12 2014
    freeman.wang@gmail.com wrote:
    I'm trying to figure out what is causing iperf UDP test to some kind
    of 'flow controlled' with the new Fedora 18/Kernel 3.x.

    I tested with the same iperf v2.0.5, same command on the client side:

    iperf -c _server_ip_addr -u -b 4000m -i 1

    With Fedora 14, the client side shows 4Gbps of packets are pushed to
    the driver and the server side shows huge packet loss. With Fedora
    18, the client side seems automatically adjusts its TX rate to the
    capacity of the link and the server side shows almost 0 packet loss.

    I'm very curious what is changed between Fedora 14 and Fedora 18 to
    give UDP such a different behaviour, and is it the future of UDP implementation?

    My experience with netperf and its UDP_STREAM test is that Linux has
    had "intra stack" flow control for a Very Long Time (tm). If the
    process tried to send outbound UDP traffic faster than the link could
    send it, it would get flow-controlled. From time to time there may
    have been bugs in that where it might get broken of course...

    rick jones
    --
    Wisdom Teeth are impacted, people are affected by the effects of events.
    these opinions are mine, all mine; HP might not want them anyway... :)
    feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

    --- MBSE BBS v1.0.1 (GNU/Linux-i386)
    * Origin: the Unofficial HP (110:110/2002@linuxnet)