[Dxspider-support] Users vs Performance?

[Kelly Jones]

Hi Arnold and the rest of the list...

Yes, I'll be happy to give everyone a little more info

>Could you provide us with some more info, like kernel version, perl
>version and what is the log telling you on the moment the "hang" appears?
>Specially the last lines of /spider/data/debug/2003/0xx.dat would be
>handy (if there is something at all)

While watching the log with watchdbg, 2 of the 4 times the last line was 
when the cron executed.  I currently on have 1 line in the file which is a 
connect KE9KD-2 if not connected statement.  KE9KD-2 is my current CLX box 
that is being retired, but I still want to feed it spots until it goes 
completely QRT.

The other two times the console.pl hung there didn't seem to be anything 
unusual.  I seem to recall one of the two just showing a PC17 frame being 
sent to my link partners, the 1st time it happened, I didn't think to look 
in the log thinking I caused it to hang while messing around.

>a perl -v tells you with perl version you have. I have seen that the
>latest versions of perl are build thread-multi with some distros,
>perhaps this could be the cause? Although I doubt that Slackware 7.1
>would carry such version of perl.

The perl version may be the issue.  Doing a perl -v I get:  This is perl 
v5.6.1 built for i386-linux.  Maybe upgrading to 5.8.0 multi-thread would 
help?  If I understand this correctly, my current version must be single 
threaded which would mean all process have to "stand in line" to get 
processed.  Perhaps a multi-thread would sove the problem?

Something else I noticed late last night (2am ;-) ) while messing with 
this, is that IF I let everything alone, Spider recovers.  It seemed to 
take anywhere from ~30 seconds to a few minutes.  This would lead me to 
believe some sort of "queue" is getting backed up and once it settles down, 
things return to normal.

I will upgrade my perl version and see what I can break doing that. ;-)

The kernel is 2.4.5 custom compile for this box with AX25 in the kernel.

Thanks for everyone's suggestions.  I'll let you know the results.

Kelly - KE9KD
http://www.dx-central.com


>sysop at wrk:/spider/perl> perl -v
>
>This is perl, v5.8.0 built for i586-linux-thread-multi
>
>Copyright 1987-2002, Larry Wall
>
>I have done in the past some tests too with a lot of clients connected
>and I could easily reach 200 (I did't continue beyond that point)
>without any problem.
>
> >
> > It seems that once I get ~40 users logged in the console hangs.  Also, 
> upon
> > login attempt, after entering your callsign, nothing happens.  The whole
> > cluster part comes to a screeching halt.
> >
> > I was able to verify this by logging in 60 "robots" and waiting for Spider
> > to break.  Sure enough, I can duplicate this "hang" on demand.
> >
> > Is there a functional limitation on users/node connections?  It's not
> > uncommon for there to be upwards of 75 users on my node.  If anyone 
> needs a
> > demonstration of the "hanging" symptoms, I'll be happy to oblige.
> >
> > The system is a Celeron 500, 256 meg RAM, ATA100 drive, Slackware
> > 7.1.  Also updated Spider to 1.51 b 57.191.
> >
> > Thanks for any ideas,
> > Kelly - KE9KD
> >
> >
> > _______________________________________________
> > Dxspider-support mailing list
> > Dxspider-support at dxcluster.org
> > http://www.tobit.co.uk/mailman/listinfo/dxspider-support
>
>--
>'73 Arnold (pa4ab / ea1cyz)
>
>e-mail: <Pa4ab at pa4ab.net>
>
>
>_______________________________________________
>Dxspider-support mailing list
>Dxspider-support at dxcluster.org
>http://www.tobit.co.uk/mailman/listinfo/dxspider-support