[Xastir-dev] Xastir a CPU pig again?

[Tom Russo]

On Thu, Dec 25, 2003 at 04:20:38PM -0500, a Mr. Richard Feyler of Fort Lee, New Jersey <bdheaton at c4i2.com> writes 'Dear Rosanne Rosannadanna':
> Tom,
> 
> 	I've got current CVS and gdal CVS running as we speak on a mini-itx
> system (533Mhz Via Eden).  Xastir is currently connected to my Igate
> taking a full feed and displaying a Tigermap.  Its showing between .7-2%
> CPU usage.

You are probably running one of the versions of Linux that never showed this
sort of performance issue even before the nexttime=2 fix, am I right?

AFAIK, Jack Twilley is the one of the few xastir users who uses FreeBSD as I
do.  I'm curious if he sees a similar issue.

> 	Just as an experiment could you shutdown the TNC interface and connect
> to an internet server?  This may help folks isolate whether the issue is
> in the TNC code or somwhere else. 

Yeah, sure.  I ran the 8 Dec version with TNC enabled linked against gdal-1.1.9
and as soon as the TNC initializion was finished the CPU time went wild.  I 
shut down that interface and enabled a heavily-filtered internet server (has a 
buddy list and a r/35/-106/375 filter rule) instead.  CPU usage was unchanged.
I then disabled "activate on startup" for my TNC, exited Xastir and started up 
with no interface enabled.  CPU was down in the noise until I turned on the 
internet server interface again, at which point it once again went up to 70%.  
If I went into Start/Stop and stopped all interfaces, CPU usage dropped back 
down into the noise.  

So it is clearly an interface issue, but not specific to the TNC code.

If I back up to a version of xastir checked out with -D "Dec 4, 2003" the
CPU usage goes back down to .2%.

> On Thu, 2003-12-25 at 15:12, Tom Russo wrote:
> > On Thu, Dec 25, 2003 at 11:53:29AM -0700, a Mr. Richard Feyler of Fort Lee, New Jersey <russo at bogoflux.losalamos.nm.us> writes 'Dear Rosanne Rosannadanna':
> > > After the "nexttime=2" fix back in late November I got too busy to mess with
> > > xastir anymore, and hadn't rebuilt from CVS since then.  For the whole
> > > time since then, xastir has taken less than 1% of the CPU except when it's 
> > > reading maps.  I built last night, and all of a sudden Xastir is back to 
> > > taking up in excess of 70% of the CPU time.
> > 
> > BTW, this is on FreeBSD, a 2.6 GHz processor, and with a KPC3+ TNC.
> > 
> > > Has anyone else who saw the CPU burning up before the "nexttime=2" fix seen 
> > > this happen?  Anyone have a clue what has changed since November 29th that 
> > > would have caused such a massive performance hit?
> > 
> > I did a bunch of cvs checkout -D "..."'s to try to find when this happened.
> > 
> > I built the version from 1 Dec, and found it *also* burned up the CPU.  The 
> > previous version I'd had built on 29 Nov didn't.  
> > 
> > Turns out that building the 1 Dec version against the CVS version of gdal
> > was the problem here --- until last night the only version of gdal I had on my
> > system was 1.1.9, the last release.  I had installed the new cvs version of
> > gdal to start playing with the new OGR-enabled xastir code.  When I removed 
> > the new CVS version of gdal and re-installed version 1.1.9, the 1 Dec version 
> > of Xastir stopped eating CPU.  It's now running at 0.1-0.2%.  Guess I am 
> > stuck without cutting edge gdal features for a while.  All other testing was
> > done with gdal-1.1.9 to eliminate that as an issue.
> > 
> > Backing out the GDAL version does not fix the current CVS version of xastir, 
> > though.  Even built against gdal-1.1.9, it still eats CPU at a rate of 70%.  
> > So, too, does a version from 8 Dec.   But the version from 4 Dec does *not* 
> > do this.
> > 
> > Anyone have an idea where to look?  Clearly it's a change that happened 
> > between 4 and 8 Dec.  
> > 
> > I briefly suspected an inter-character pacing issue, as the versions earlier
> > than 8 Dec take a great deal longer to get through the TNC interface 
> > initialization strings than either the 8 Dec or 25 Dec versions, but the 
> > timing slider seems to have no effect.  Prior to the implementation of the
> > pacing slider, what was the inter-char delay?
> > 
> > I did notice, however, that the ballooning of CPU usage occurs as soon as
> > the initialization of the TNC is done and the first posit is transmitted.  
> > Until that moment the CPU usage is below 20% and drifting down.
> > 
> > BTW, in trying all this, I had tried to configure xastir with "--without-gdal"
> > only to find that there is no such option recognized.  xastir's configure
> > will build against any installed gdal and not allow configure-time disabling
> > of it.  

-- 
Tom Russo    KM5VY   SAR502   DM64ux          http://www.swcp.com/~russo/
Tijeras, NM  QRPL#1592 K2#398  SOC#236 AHTB#1 http://www.qsl.net/~km5vy/
 echo "prpv_a'rfg_cnf_har_cvcr" | sed -e 's/_/ /g' | tr [a-m][n-z] [n-z][a-m]