// set up TX streams and threads
std::thread tx_threads[NUM_CHANNELS];
uhd::tx_streamer::sptr tx_streams[NUM_CHANNELS];
uhd::tx_metadata_t md;
md.start_of_burst = true;
md.end_of_burst = false;
md.has_time_spec = true;
md.time_spec = uhd::time_spec_t(0.1);
for (int i = 0; i < NUM_CHANNELS; i++){
stream_args.channels = std::vector<size_t>(1,i);
tx_streams[i] = usrp->get_tx_stream(stream_args);
//start the thread
std::cout << "Starting tx thread " << i << std::endl;
tx_threads[i] = std::thread(txTask,tx_buffs[i],tx_streams[i],md);
}
///////////////////////////////////////////////////
////////the thread function////////////////////////
////////////////////////////////////////////////////
void txTask(Complex *buff, uhd::tx_streamer::sptr tx_stream,
uhd::tx_metadata_t md){
size_t num_acc_samps = 0;
struct timespec start_time;
clock_gettime(CLOCK_MONOTONIC, &start_time);
while(!stop_signal_called){
size_t samples_sent = tx_stream->send(buff,BUFF_SIZE,md);
num_acc_samps += samples_sent;
md.start_of_burst = false; //after first transmission set
to false
md.has_time_spec = false; //after first transmission set to false
}
struct timespec end_time;
clock_gettime(CLOCK_MONOTONIC, &end_time);
double runtime = (end_time.tv_sec - start_time.tv_sec) +
(end_time.tv_nsec - start_time.tv_nsec ) /
1000000000.0;
std::cout << std::endl << "Sent " << num_acc_samps
<< " samples in " << runtime << "s"
<< " Throughput = " << num_acc_samps / 1e6 /runtime << "
Msps"
<< std::endl;
//send a mini EOB packet
md.end_of_burst = true;
tx_stream -> send("",0,md);
printf("End transmit \n");
}
Even with the metadata fix I still get underflows when transmitting
and receiving on the same channel, no matter what the rate. If you
have an X310, I wonder if you could compile and run my code and see if
you get the same result?
Thanks,
Jason
------------------------------------------------------------------------
*From:* The Tilla <ti...@comcast.net>
*Sent:* Monday, August 14, 2017 4:47:31 PM
*To:* Jason W Zheng; mle...@ripnet.com
*Cc:* usrp-users@lists.ettus.com
*Subject:* RE: [USRP-users] Buffer underrun issue with simultaneous
transmit and receive on the X310
I think some of the issue here as well is the tx_metadata.
For continuous streaming, the first packet ONLY should have
startofburst = true…
Further packets after the first one should have startofburst = false,
cuz they really are not start of burst, they are continuations of the
initial burst.
Furthermore, for continuous streaming as soon as possible, you should
not provide a timespec.
If you continually provide a timespec to each packet, it will
continually try to send them at time = 0.1, which is not what you want…
If you want to start your tx at a specific time, then similar to
startofburst, you should only provide a timespec with the first
packet, all subsequent packets should not include a timespec.
Does that make sense?
It can be a bit confusing at first, but after writing a few apps, you
will be an expert 😊
*From:* Jason W Zheng [mailto:jason.w.zh...@aero.org]
*Sent:* Monday, August 14, 2017 4:41 PM
*To:* ROBIN TORTORA <ti...@comcast.net>; mle...@ripnet.com
*Cc:* usrp-users@lists.ettus.com
*Subject:* Re: [USRP-users] Buffer underrun issue with simultaneous
transmit and receive on the X310
Hi Robin, Marcus
It makes no sense to me, then why changing transmit to be on its own
channel (daughtercard) would cause the program to work.
// set up TX streams and threads
std::thread tx_threads[NUM_CHANNELS];
uhd::tx_streamer::sptr tx_streams[NUM_CHANNELS];
uhd::tx_metadata_t md;
md.start_of_burst = true;
md.end_of_burst = false;
md.has_time_spec = true;
md.time_spec = uhd::time_spec_t(0.1);
for (int i = 0; i < NUM_CHANNELS; i++){
stream_args.channels = std::vector<size_t>(1,i); //changing this from
i to 1 causes it to work for unknown reasons
tx_streams[i] = usrp->get_tx_stream(stream_args);
//start the thread
std::cout << "Starting tx thread " << i << std::endl;
tx_threads[i] = std::thread(txTask,tx_buffs[i],tx_streams[i],md);
}
The relevant line is commented in the above code snippet. When I
change the channels arguments for the tx stream_args to be on it's own
channel (daughterboard) separate from the receive, the underflows go
away.
@Marcus, in the code I posted, there is no coordination between the
transmit and the receive threads. This is for example purposes. The
transmit and receive buffers are independent. The transmit thread
buffer is prefilled with 0's and is constantly transmitting 0's.
------------------------------------------------------------------------
*From:*ROBIN TORTORA <ti...@comcast.net <mailto:ti...@comcast.net>>
*Sent:* Monday, August 14, 2017 1:06:20 PM
*To:* Jason W Zheng via USRP-users; Jason W Zheng
*Subject:* Re: [USRP-users] Buffer underrun issue with simultaneous
transmit and receive on the X310
Willing to bet significant money your tx thread is being put on the
physical processor that does not have your NIC card attached to it...
Then, all data must go over QPI between processors.
NUMA is not good for this.
You can experiment with affinity to make sure things are all on the
same physical processor as your NIC...
We no longer buy multi-processor boxes for issues like this, NUMA is
bad...
On August 14, 2017 at 2:32 PM Jason W Zheng via USRP-users
<usrp-users@lists.ettus.com <mailto:usrp-users@lists.ettus.com>>
wrote:
I'm running on a server with 2 Intel Xeon E5-2650 v4, CPU is
listed here:
https://ark.intel.com/products/91767/Intel-Xeon-Processor-E5-2650-v4-30M-Cache-2_20-GHz
The network card I'm using is the recommended Intel X520-DA2
I also have over 500 gigs of ram, so that is not an issue
I don't think my cpu or network card are the issue as when I
change my code to receive and transmit on separate channels, it
works without any underflows.
------------------------------------------------------------------------
*From:*USRP-users <usrp-users-boun...@lists.ettus.com
<mailto:usrp-users-boun...@lists.ettus.com>> on behalf of Jason W
Zheng via USRP-users <usrp-users@lists.ettus.com
<mailto:usrp-users@lists.ettus.com>>
*Sent:* Monday, August 14, 2017 11:27:24 AM
*To:* ROBIN TORTORA
*Cc:* usrp-users@lists.ettus.com <mailto:usrp-users@lists.ettus.com>
*Subject:* Re: [USRP-users] Buffer underrun issue with
simultaneous transmit and receive on the X310
How should I set the tx metadata? My application will essentially
stream continuously.
Thanks,
Jason
------------------------------------------------------------------------
*From:*ROBIN TORTORA <ti...@comcast.net <mailto:ti...@comcast.net>>
*Sent:* Monday, August 14, 2017 11:17:07 AM
*To:* Jason W Zheng via USRP-users; Jason W Zheng
*Subject:* Re: [USRP-users] Buffer underrun issue with
simultaneous transmit and receive on the X310
I dont see you setting the tx metadata object members to any
values, so you are essentially going to tx with default metadata
IF there is a constructor that initializes all the members to a
consistent value...
On August 14, 2017 at 1:56 PM Jason W Zheng via USRP-users
<usrp-users@lists.ettus.com
<mailto:usrp-users@lists.ettus.com>> wrote:
Since I haven't gotten a response in a week, I thought some
source code might help. I've removed any buffer dependencies
leaving just the receive thread and transmit thread. The
receive thread is constant receiving to a buffer, while the
transmit thread is constantly transmitting 0s from another
buffer.
On the x310, when I receive and transmit from the same channel
(same daughterboard), I get underflows (U's on the console),
no matter what the rate. When I put receive and transmit on
separate channels, the U's go away.
How do I fix my source code so I can receive and transmit from
the same channel without underflows?
Thanks,
Jason
------------------------------------------------------------------------
*From:*USRP-users <usrp-users-boun...@lists.ettus.com
<mailto:usrp-users-boun...@lists.ettus.com>> on behalf of
Jason W Zheng via USRP-users <usrp-users@lists.ettus.com
<mailto:usrp-users@lists.ettus.com>>
*Sent:* Monday, August 7, 2017 1:05:05 PM
*To:* usrp-users@lists.ettus.com
<mailto:usrp-users@lists.ettus.com>
*Subject:* [USRP-users] Buffer underrun issue with
simultaneous transmit and receive on the X310
Hi,
I'm building an application where I receive data from the
x310, process the data, then transmit it out. I have 3
separate threads running, one for each task. The receive task
is constantly receiving data from the x310 at 200MS/s and
putting it into a buffer. The process task takes data from the
receive buffer, processes it, and puts it into a transmit
buffer. The transmit task is constantly transmitting data from
the transmit buffer at 12.5 MS/s.
I get constant underruns (U's on the console) when I try to
receive and transmit from the same channel on the x310.
However, when I transmit and receive on separate channels,
underruns no longer occur. I would like to figure out a
solution to this problem as I want to processes two 200MS/s
streams on 1 x310.
This problem occurs no matter the sample rates. I've tried
lowering the receive rate to 50MS/s (while transmitting 4x the
data to balance things out) and there are still constant
underruns though at a lower rate. I know processing time is
not the issue as I've profiled the time it takes to process,
and I have even removed the processing task from the
application altogether (receive data to a buffer and do
nothing with the data; constantly transmit 0's from the
transmit buffer) and the underruns still occur.
I'm running on UHD 3.11.0, and the x310 is is configured with
basic TX/RX daughterboards and flashed with the XG image for
use over two 10 gig ethernet links.
Thanks,
Jason
_______________________________________________
USRP-users mailing list
USRP-users@lists.ettus.com <mailto:USRP-users@lists.ettus.com>
http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com
_______________________________________________
USRP-users mailing list
USRP-users@lists.ettus.com <mailto:USRP-users@lists.ettus.com>
http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com
_______________________________________________
USRP-users mailing list
USRP-users@lists.ettus.com
http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com
