On Fri, 7 Nov 2014, e1000-devel-requ...@lists.sourceforge.net wrote:
> Date: Thu, 6 Nov 2014 14:31:14 +0100
> From: Julia Niewiejska <julia.niewiej...@fkie.fraunhofer.de>
> Subject: Re: [E1000-devel] 802.3x pause frames
>
> Hi,
>
>
> Ping was suggested in [1] as a test for whether it's working. I also
> tried saturating the link with UDP packets in the setup with 2 physical
> machines where the link speed was set to 10 Mbps. I used iperf for this
> purpose, started the transmission from first to second machine, and
> simulaneously used the tool from [1] to generate a couple of pause
> frames at arbitrary times on the second machine. I captured the traffic
> at the second machine and considered the timestamps.
>
> According to the standard, the network adapter pauses the transmission
> completely on the reception of a pause frame for the time specified.
> Therefore I would expect to see noticeable variations in the intervals
> between consecutive packets at the time when I generated the pause
> frames. However there were no significant variations to be seen.
>
>
> Regards,
>
> Julia Niewiejska
>
>
> [1] http://www.tux.org/pub/sites/www.zip.com.au/%257Eakpm/linux/#flow-ctrl
Julia-
I am not sure how you are capturing your traffic on the second machine,
but if you are not using hardware timestamping, be extremely cautious with
the timestamps you receive. If you are using libpcap without hardware
timestamps, you are relying on the kernel for the reception time and
normally NICs will "bundle" packets and hand them up at one time, leading
to discretized delta-t's between the packets in your capture. This may
mask the variations you are looking for.
Best of luck,
Steve
------------------------------------------------------------------------------
_______________________________________________
E1000-devel mailing list
E1000-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/e1000-devel
To learn more about Intel® Ethernet, visit
http://communities.intel.com/community/wired
