On Tue, Mar 9, 2021 at 10:03 PM Doug Blackburn via USRP-users <
usrp-users@lists.ettus.com> wrote:

> Hello --
>
> I've got some questions re: latency with the x300 over the 10GigE
> interface.
>
> If I use the latency_test example operating at a rate of 50 MSPS, I have
> no issues with a latency of 1ms.  The latency test receives data, examines
> the time stamp, and transmits a single packet.
>
> I have an application where I'd like to run the transmitter continuously,
> and I got curious about the latency involved in that operation.  My
> application is similar to the benchmark_rate example.  I added the
> following lines to the benchmark_rate example at line 256 after the line.
>
> md.has_time_spec = false;
>
> ====
> if ( (num_tx_samps % 50000000) < 4*max_samps_per_packet )
> {
>     uhd::time_spec_t expectedTime = startTime + (double) ( num_tx_samps  )
>                       / (double)usrp->get_tx_rate();
>     uhd::time_spec_t timeAtLog = usrp->get_time_now();
>     timeAtLog = usrp->get_time_now();
>     std::cerr << "==== Actual time ====" << std::endl;
>     std::cerr << "     " << timeAtLog.get_full_secs() << " / "
>                           << timeAtLog.get_frac_secs() << std::endl;
>     std::cerr << "==== Expected time ====" << std::endl;
>     std::cerr << "     " << expectedTime.get_full_secs() << " / "
>                           << expectedTime.get_frac_secs() << std::endl;
> }
> ====
>
> The intent of this insertion is to log the time at which we return from
> tx_stream->send() and the time at which the first sample of that sent data
> should be transmitted -- at approximately once per second when running at
> 50 MSPS.
>
> After the first second, I consistently saw the following results:
>
> ==== Actual time ====
>      1 / 0.10517
> ==== Expected time ====
>      1 / 0.27253
>
> ==== Actual time ====
>      1 / 0.105419
> ==== Expected time ====
>      1 / 0.27255
>
> Which indicates to me that there is a latency of approximately 167ms when
> transmitting data.  That is, the send() function is returning 167ms earlier
> than I expect the data to actually be transmitted.   If I halve the sample
> rate to 25 MSPS, the latency doubles.
>
> What is the source of the latency when running in a continuous mode?
> Initially, I had thought that this might be related to the
> send_buffer_size, but making changes to send_buffer_size seem to not have
> an effect.   FWIW, 167ms at 50 MSPS is suspiciously close to the value for
> wmem_max (33554432) suggested in the x300 system configuration ... but
> neither changing that value or send_buffer_size seems to make a difference.
>
> Is this latency tunable?
>

I suspect it's the DMA FIFO which uses the DRAM in the X310 as a buffer.
The default buffer size is 32MB.

Which version of UHD are you using?

Brian
_______________________________________________
USRP-users mailing list
USRP-users@lists.ettus.com
http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com

Reply via email to