Re: [Discuss-gnuradio] dropped bits?

[Eric Blossom]

On Sat, Jul 02, 2005 at 10:57:42PM -0700, John E. Don Carlos wrote:
> I added head followed by save_file to tvrx_debug.py to save data for further
> analysis.  The terminal shows some u0u0 activity which may indicate that
> data is lost.  tvrx_debug does not output u0u0 without save_file (with or
> without head) unless the mouse is moved.  Apparently mouse motion or
> save_file is interfering with the USB transfer, perhaps delaying servicing
> USB interrupts?  The decimated sample rate is 8MSamples/sec.

The problem is probably related to disk throughput, especially if
you're on a laptop.  We do need to figure out a way to enable the real
time scheduler, or at least enough negative niceness to keep us at a
better priority than the X-server.  Last time I checked, there is a
bit of subtlety involved given our use of multiple threads.

> Is there a way to store the data sample in memory and save it to file at the
> end of the data gathering period?

You could use gr.head and a gr.vector_sink_{s,f,c}.  When the graph was
done running [fg.run()], you could grab sink.data() and then write it
to disk.

> I don't know USB or the USRP details, but at 32MB/s an interupt would occur
> every 15 ms if the data is buffered in 0.5 MB chunks.  My pentium is
> hyper-threaded - is there a way to dedicate one thread to service the USB
> interupts?

It's not missing interrupts.  The user mode application isn't keeping
up most likely because of disk i/o and because we're scheduled at a
lower pri than the X-server.

Eric