[netperf-dev] [PATCH] Avoid a double htonl in udp_stream results

[David MacMahon]

On Jul 21, 2011, at 1:37 PM, Rick Jones wrote:

> Do you have some before/after output showing the number wrong and then correct? (Yes, I'm being a paranoid pedant :)

The netperf executable (in send_udp_stream; not sure about the omni tests) ignores the bytes_received field of the results "packet" and instead computes it from (send_size * messages_recvd), so this change has no effect on the output of netperf.

I am using netserver, but not netperf, to test transfer rates from an FPGA sending UDP packets over 10 GbE to a PC running netserver.  I have a small script that establishes the control connection to netserver to select the test (UDP_STREAM) and provide parameters.  It uses the response "packet" to tell the FPGA which socket to use, waits for the test to finish, then reads the results "packet".  I could also compute bytes_received from (send_size * messages_received), but that assumes all messages are send_size which could mask errors on the FPGA side (e.g. smaller packets).

FWIW, I'm also seeing an off-by-one error in bytes_received that I think it is due to adding the -1 "length" returned be recv to bytes_received, but I haven't fully tracked that down yet (with UDP_STREAM test).

Dave