[Discuss-gnuradio] TX Flow Control acting weird on uhd

[Gaetano Mendola]

Hi,
I'm having an issue and the culprit seems the TX Flow Control
implemented in the UHD
(got it from GIT yesterday).

I have a very powerful server, dual socket Xeon with 6 cores per
socket, 12 GB Ram 1333Mhz.
I have already another thread about this issue on this mailing list,
I'm starting another thread
with the new info I have got during last week of digging into this very problem.

My application has to send samples with a rate of 12.5 Mega samples
per second, it sends
those samples in chunks of 45000 samples. This means that each send
has to last around
0.0036 seconds.
When things goes well I obtain indeed an average of 0.0036 secs, due
to some internal buffer in
the N210 device some time the send execution time is lower than 0.0036
and some time is higher
as you can see in the attached image send_response_time.gif.
However most of the time the execution time is the one shown in the
corrupted_tx.gif, as you can
see the execution times have some accumulation values multiple of around 0.1.
Consider same application connected with an old USRP2 using then the
libusrp2 doesn't have any
problem at all. The same UHD doesn't have any problem on other old servers.

Josh,
I can give you access to our server if you want instrument somehow the
UHD to see what is going
on.

Regards
Gaetano Mendola

-- 
cpp-today.blogspot.com

send_response_time.gif
Description: GIF image

corrupted_tx.gif
Description: GIF image