Re: [Discuss-gnuradio] USRP2 spontaneously dies while transmitting

[Eric Blossom]

On Tue, Aug 04, 2009 at 07:44:21PM -0400, Andrew Lutomirski wrote:
> On Tue, Aug 4, 2009 at 6:05 PM, Eric Blossom<address@hidden> wrote:
> >
> > You're running it through a switch.  We tell you not to do that.
> > Your switch is probably declining to be flow controlled by the USRP2,
> > hence there's no back pressure being applied to the host, thus the
> > unbounded amount of data sent from the host.
> >
> > Rule 1: Don't connect your USRP2 via a switch
> >
> > Rule 2: If you think you know what you're doing and understand the ins
> >        and outs of ethernet PAUSE frames and asymmetric flow control,
> >        and can get your switch to accept flow control from the USRP,
> >        then it _may_ be possible to make it work through a switch.
> >
> > Rule 3: See Rule 1.
> 
> Would you accept a patch to error out on send if sustained data rates
> more than, say, 150% of what they should be are seen?

I'd definitely consider one.  Be sure it doesn't burn many cpu cycles.
When running with interp = 4, even on big iron, we don't have much to spare.

> Also, I fully expect (and don't mind) some drops due to running
> through a switch.  I'm only running this thing when the network is
> otherwise nearly idle and even then I'm at high interpolation.  It
> still seems buggy that the USRP2 crashes as a result.

I agree that it shouldn't crash or disappear.  Feel free to debug and
fix it.

> Is the USRP2 ethernet protocol documented somewhere?  Quick googling
> didn't find it.

The existing protocol was stopgap kludge that escaped into the wild.
The packet formats are defined in usrp2/firmware/include/usrp2_eth_packet.h

The entire on-the-wire protocol will be getting reworked to use VRT.

Eric