[Xastir-dev] Xastir a CPU pig again?

[Brian D Heaton]

Tom,

	Hmm, hopefully Curt will be back in a day or two and can let us know
what the rationale was behind the change and if flipping it to 1ms will
work for everyone.

	I think you're right about the pthreads library.  I beat my head on it
a while back for some work on aprsd.  I wonder if BSD and some versions
of Linux ship with a different version than I've been using.  Likely it
has something to do with the efficiency of acquiring and holding the
mutex locks during the send operation.  

	In the good new department I got my Igate/digi running again.  A little
antenna work tomorrow and all will be well here.  I'll probably also
swap out the old TR-7800 that is currently the digi/igate radio with an
FT-2600 I just picked up.  Now I've just got to get comfy with digi_ned
so I can do more intelligent stuff than whats available in the KPC3+.
<G>

			73/N5VFF



On Thu, 2003-12-25 at 19:33, Tom Russo wrote:
> On Thu, Dec 25, 2003 at 05:16:51PM -0700, a Mr. Richard Feyler of Fort Lee, New Jersey <russo at bogoflux.losalamos.nm.us> writes 'Dear Rosanne Rosannadanna':
> > On Thu, Dec 25, 2003 at 06:08:33PM -0500, a Mr. Richard Feyler of Fort Lee, New Jersey <bdheaton at c4i2.com> writes 'Dear Rosanne Rosannadanna':
> > > On Thu, 2003-12-25 at 18:05, Tom Russo wrote:
> > > > I can confirm it is in fact exactly the source of the problem.  I
> > > > changed the 100us delay in interface.c that had been reduced from
> > > > 200ms sometime between 4 Dec and 8 Dec.  In the 25 Dec CVS version
> > > > this is on line 5307.  Setting it to 100ms instead of 100us ended my
> > > > problems.  With that change (and no other) both the 8 Dec and current
> > > > CVS versions of Xastir are back down to taking negligible CPU times.
> > 
> > > > Now to figure out if the issue with CVS gdal instead of gdal-1.1.9
> > > > is a real one, or just a red herring.
> > 
> > > 
> > > 	Interesting.  I wonder if th 100us vs. 100ms was a typo when the change
> > > was committed? 
> > 
> > Well, the tmv.tv_usec setting went from 200000 to 100 in one edit, and
> > the units in the comments changed to match it as well.  I'm guessin'
> > it was intentional and related to the change that resulted in the
> > inter-char pacing slider.
> 
> Actually, the commit log shows that it was indeed intentional:
> 
> -------------------------
> Revision 1.146
> Sun Dec 7 20:23:28 2003 UTC (2 weeks, 4 days ago) by we7u
> Changes since 1.145: +4 -3 lines
> 
> Reducing delay in port_write for checking queue.  Down to 100us from 200ms.
> Makes writing to interfaces much snappier, but doesn't bog down CPU.
> Cranking up channel_data() busy wait from 1 to 2us.
> 
> -----------------------
> 
> I just finished a quick study in reducing the 100ms delay down to 10ms
> and then 1ms.  No CPU bogging here.  Drop it down one more decade to 100us 
> and kaboom.