System crashing when swap hits 25%

[Simon Slater]

On Wed, 2008-03-12 at 12:22 +1100, Simon Slater wrote:
> On Tue, 2008-03-11 at 12:53 -0400, Chris Snook wrote:
> > Simon Slater wrote:
> > > On Mon, 2008-03-10 at 21:01 -0400, Chris Snook wrote:
> > >>>> sysrq-m and sysrq-t data out of the system.  You'll want to set 
> > >>>> kernel.sysrq=1 first. 
> > >>> Where do I find these?
> > >>>
> > >> There are two interfaces.  One is the /proc/sys directory, and the
> > >> other 
> > >> is sysctl.  If you want to set something with sysctl, the easy way is
> > >> to 
> > >> add a line in sysctl.conf (for example: vm.overcommit_memory=2) and
> > >> then 
> > >> run sysctl -p to process the config file again.
> > >>
> > > 	Have just been through a few man pages.
> > > Temporarily, the command sysctl vm.overcommit_memory=2 would do the job?
> > 
> > By itself, probably not.
> > 
> > > What does the value '2' signify?
> > 
> > 0 = heuristic overcommit
> > 1 = always overcommit (until completely out of RAM + swap)
> > 2 = overcommit (swap + (vm.overcommit_ratio * RAM))
> > 
> > > What do sysrq-m and sysrq-t do?  I haven't found them yet.
> > 
> > alt+sysrq+m prints memory statistics
> > alt+sysrq+t prints thread state information
> 
> Silly me. It's been so long since I used that key I forgot it is there.
> > 
> > > Oh, currently vm.overcommit_memory=0 and kernel.sysrq=0.
> > 
> > You'll need to set kernel.sysrq=1 for keyboard sysrq input to be 
> > honored.  If you set vm.overcommit_memory=2, then vm.overcommit_ratio 
> > will be honored.
> > 
> > -- Chris
> > 
> 	I used sysctl vm.overcommit_memory=2 and checked for badblocks in the
> swap file (none found) and performance is amazing.  This is very
> subjective, but I can load and run so much more and I'm sure things are
> runnung faster.
> 
> 	At the moment the panel applet is at the point where things usually
> hang.  free reports:
> [simon at Ipex ~]$ free
>              total       used       free     shared    buffers
> cached
> Mem:        515220     509020       6200          0       5596
> 215728
> -/+ buffers/cache:     287696     227524
> Swap:      1048568     213904     834664
> However, rather than the usual 2 or 3 apps with a few browser tabs open
> when things hang, at the moment I have: terminal; Firefox with 20 tabs;
> 2x Konqueror one with 3 tabs; Evolution downloading email and
> spamassassin; 24 OOWriter docs; 6 OOImpress presentations; 11 Kpdf with
> 11Mb pdf and 1 10Mb pdf; 13 brochures open in Evince; 2 tty sessions
> with 1 running an smbtar script and jobs to the printer.  This is
> amazing.
> 
> 	I'll push things further now to see what happens as swap increases.

	Here's an update.  From this point I loaded some web sites with Flash &
Java and things really started to slow down and KDE alone began to crash
repeatedly.  I supposed a PII can only take so much so began closing the
PDFs and OO stuff.  When the swap got down to below 100000 the system
hung again.

	So I started thinking perhaps it was the Java or Flash.  I used the
Live Timing from www.formula1.com during the practice and qualifying
sessions on Friday & Saturday for the Melbourne Grand Prix to test the
idea.  The system would hang, but not in what I could find to be a
repeatable pattern.  While watching the race on Sunday I was printing
form letters and had the live timing up.  Every time my wife started the
vacuum cleaner the system hung!

	I'll check the power supply voltages over Easter, since this looks most
likely. I won't have time till then, need to do an engine change in our
van before Easter.  We have had a week of 40+ degree days (Celsius) so
maybe the town supply has been stretched too.

	Sorry if I led everyone on a wild goose chase.  Did learn some more
though and have more to do homework on. Thanks again.

-- 
Regards
Simon