Empathy List Archives

JL

Joe Leech

Tue, Mar 3, 2015 10:20 PM

Hi everyone,

I ran into a RX timeout issue where if I don't issue a stream cmd within a certain amount of time, the RX side times out whenever I do a recv(). I looked through the USRP mailing list and saw that someone else reported the same problem about half a year ago here: http://lists.ettus.com/pipermail/usrp-users_lists.ettus.com/2014-April/009329.html

But for me, I can reproduce the timeout behavior after just 30 seconds of inactivity, instead of 64 as noted in the other thread.

My setup:
Linux 3.12.26 based on Debian
UHD 003.008.002-86-g566dbc2b
x300 box with SBX-120 and 10GbE NIC recommended on Ettus website

Test script:
//initialization routine... pretty generic like make, subdev spec, etc.

//boost::this_thread::sleep(boost::posix_time::seconds(30)); //NOTE: this sleep breaks recv()
//RX stream_cmd_t
stream_cmd_t stream_cmd(stream_cmd_t::STREAM_MODE_NUM_SAMPS_AND_DONE);
stream_cmd.stream_now = false;
stream_cmd.num_samps = 10000;
stream_cmd.time_spec = usrp->get_time_now() + time_spec_t(0.2); //padded buffer time
usrp->issue_stream_cmd(stream_cmd);

rx_metadata_t md;
size_t numSampsRet = 0;
numSampsRet = rx_stream->recv(&Buff.front(), stream_cmd.num_samps, md, 5);

So like the other thread, I noticed if I initialized the USRP but I don't call issue_stream_cmd within 30 seconds, I cannot get any sample returned back to me via recv(). It looks like from the other thread that Michael West from Ettus confirmed the bug. I guess I can have a loop that issues a stream cmd every 29 seconds to
prevent the RX from timing out, but it is not really an ideal solution
and adds unnecessary network traffic. So I am wondering if there is any estimated time or UHD driver version that will include the fix? Or maybe there is a better workaround? The other thread describes putting the sleep before usrp->get_rx_stream() but that doesn't really apply to me since I could have a long idle wait between just initializing the box (which includes get_rx_stream) and actually issuing a cmd to capture IQ...

As always, any help is greatly appreciated! Thanks everyone!
Joe

Hi everyone, I ran into a RX timeout issue where if I don't issue a stream cmd within a certain amount of time, the RX side times out whenever I do a recv(). I looked through the USRP mailing list and saw that someone else reported the same problem about half a year ago here: http://lists.ettus.com/pipermail/usrp-users_lists.ettus.com/2014-April/009329.html But for me, I can reproduce the timeout behavior after just 30 seconds of inactivity, instead of 64 as noted in the other thread. My setup: Linux 3.12.26 based on Debian UHD 003.008.002-86-g566dbc2b x300 box with SBX-120 and 10GbE NIC recommended on Ettus website Test script: //initialization routine... pretty generic like make, subdev spec, etc. //boost::this_thread::sleep(boost::posix_time::seconds(30)); //NOTE: this sleep breaks recv() //RX stream_cmd_t stream_cmd_t stream_cmd(stream_cmd_t::STREAM_MODE_NUM_SAMPS_AND_DONE); stream_cmd.stream_now = false; stream_cmd.num_samps = 10000; stream_cmd.time_spec = usrp->get_time_now() + time_spec_t(0.2); //padded buffer time usrp->issue_stream_cmd(stream_cmd); rx_metadata_t md; size_t numSampsRet = 0; numSampsRet = rx_stream->recv(&Buff.front(), stream_cmd.num_samps, md, 5); So like the other thread, I noticed if I initialized the USRP but I don't call issue_stream_cmd within 30 seconds, I cannot get any sample returned back to me via recv(). It looks like from the other thread that Michael West from Ettus confirmed the bug. I guess I can have a loop that issues a stream cmd every 29 seconds to prevent the RX from timing out, but it is not really an ideal solution and adds unnecessary network traffic. So I am wondering if there is any estimated time or UHD driver version that will include the fix? Or maybe there is a better workaround? The other thread describes putting the sleep before usrp->get_rx_stream() but that doesn't really apply to me since I could have a long idle wait between just initializing the box (which includes get_rx_stream) and actually issuing a cmd to capture IQ... As always, any help is greatly appreciated! Thanks everyone! Joe

JL

Joe Leech

Fri, Mar 13, 2015 10:28 PM

Hi again everyone,

Is there an update to this issue, or has no one else ran into this? Is there already a fix that I can implement to my system?

Regards,
Joe

To: usrp-users@lists.ettus.com
Date: Tue, 3 Mar 2015 14:20:47 -0800
Subject: [USRP-users] x300 RX Timeout With Issue_stream_cmd() and Recv()
From: usrp-users@lists.ettus.com