STREAMS pipes - ioctl I_SENDFD - Unix

This is a discussion on STREAMS pipes - ioctl I_SENDFD - Unix ; Rick Jones wrote: >> I don't own a NIC that will pass 12 Gigibits per Second. > But that is why netperf can present service demand - so one can > compare the efficiency of two different stacks running on ...

+ Reply to Thread
Page 2 of 2 FirstFirst 1 2
Results 21 to 22 of 22

Thread: STREAMS pipes - ioctl I_SENDFD

  1. Re: STREAMS pipes - ioctl I_SENDFD

    Rick Jones wrote:
    >> I don't own a NIC that will pass 12 Gigibits per Second.


    > But that is why netperf can present service demand - so one can
    > compare the efficiency of two different stacks running on the same
    > hardware, even if doesn't have the hardware to 12 gigabits per second
    >


    Unfortunately, to determine whether it was worth writing a STREAMS TCP,
    I was testing STREAMS UDP vs. Linux Native Sockets UDP. Both UDP_STREAM
    and XTI_UDP_STREAM tests can saturate even a GigE interface without any
    effort.

    --
    Brian F. G. Bidulock
    bidulock@openss7.org
    http://www.openss7.org/

  2. Re: STREAMS pipes - ioctl I_SENDFD

    bidulock@openss7.org wrote:
    > Rick Jones wrote:
    > >> I don't own a NIC that will pass 12 Gigibits per Second.


    > > But that is why netperf can present service demand - so one can
    > > compare the efficiency of two different stacks running on the same
    > > hardware, even if doesn't have the hardware to 12 gigabits per second
    > >


    > Unfortunately, to determine whether it was worth writing a STREAMS TCP,
    > I was testing STREAMS UDP vs. Linux Native Sockets UDP. Both UDP_STREAM
    > and XTI_UDP_STREAM tests can saturate even a GigE interface without any
    > effort.


    Again that is why netperf will report service demand if you tell it
    to measure CPU utilization, so you can see the relative efficiencies
    of the two solutions even if both saturate the NIC.

    Now, if there is no intra-stack flow-control with UDP in both cases,
    one can configure netperf with --enable-intervals and cause netperf to
    pace itself (via the itimer) to a given bitrate which you can tweak to
    be very close to say GigE saturation, and use the CPU
    utilization/service demand measurements there to get a good idea of
    relative efficiencies.

    I'm sure that works with UDP_STREAM, and ass-u-me it works with the
    XTI_UDP_STREAM (I don't get to kick the tires of the XTI tests in
    netperf as often as I might like).

    rick jones
    --
    portable adj, code that compiles under more than one compiler
    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...

+ Reply to Thread
Page 2 of 2 FirstFirst 1 2