Re: [Discuss-gnuradio] Overrun Underrun False packets issues on Gnuradio using Beagle Board

[Philip Balister]

On 04/08/2010 12:07 AM, Jeannette Nounagnon wrote:
> Hello,
>
> I have a Beagle Board running Gnuradio(3.2.1). I copied the
> gnuradio-examples folder from Gnuradio (3.2.2) on the Beagle Board (BB).
> When I run the usrp_benchmark_usb, I get a maximum throughput of
> 4MB/sec, instead of 32MB/sec on my regular Linux desktop.
> Here is my problem: On the Beagle Board, I am getting lots of dropped
> packets, most false packets, u0 and uU when running benchmark_tx/rx, and
> tunnel.py.
> My first assumption is that I have a USB bottleneck. Am I correct?

Yes. I'm not sure exactly where the bottleneck is, but I suspect the USB 
drivers for the OMAP3 have not been optimized for speed.


> I have tried to resolve these problems using a few options so far:
> Option 1. Reducing the rate (I modified the gnuradio examples files
> accordingly), so I am pretty sure I am reducing the rate of
> transmission. [The verbose statements for Requested bit rate and Actual
> bitrate are the same]. That did not resolve my problems. Why?

Benchmark_{tx,rx} add signal processing right? I would run oprofile and 
see what signal processing block is using most of the CPU time. I 
suspect that rewriting the signall processing in assembler using NEON 
will help a great deal.

Finally, I'd point out that beagleboard.org is a mentoring organization 
for Google Summer of Code. The proposal deadline is sometime tomorrow. 
We'd be glad to consider a proposal based on getting this working really 
well on the Beagle Board.


Philip


> Option 2. Changing buffer size. I am trying to tweak the buffer size(by
> modifying fusb_block_size and nblocks) , but I do not know how to check
> that the change actually occured in realtime.
> (a) Where can I find the real value of the realtime buffer size so I can
> print it in verbose. I tried to trace it down, but eventually hit a wall.
> (b) How do you think I can change the buffer size and rate in order to
> satisfy a maximum throughput of 2MB/sec and still get reliable data?
> (c) I know that buffer size is B*N, where B is a multiple of 512 bytes.
> I also know that a large N can cause latency, but I want to disregard
> any latency issues for now.
> Will I resolve anything by changing B and N? I have attempted a few
> combinations of B, N and rate, still no solutions.
>
> Any suggestions to resolve these issues are welcome.
> Thanks,
>
> Jeannette
>
>
>
> _______________________________________________
> Discuss-gnuradio mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/discuss-gnuradio