On Sat, Jan 13, 2007 at 03:30:27PM +0100, Willy Tarreau wrote:
quoted text
> On Sat, Jan 13, 2007 at 02:16:01PM +0100, Toon van der Pas wrote:
> > On Sat, Jan 13, 2007 at 08:22:18AM +0100, Willy Tarreau wrote:
> > > > 
> > > > Which makes me think that we aren't writing back fast enough.  If I  
> > > > mount the drive "sync" the issue clearly goes away.
> > > > 
> > > > It appears from an strace we are doing ftruncate64(5, 178257920) when  
> > > > we OOM.
> > > > 
> > > > Any ideas on VM parameters to tweak so we throttle this from occurring?
> > > 
> > > Take a look at /proc/sys/vm/bdflush. There are several useful parameters
> > > there (doc is in linux-xxx/Documentation). For instance, the first column
> > > is the percentage of memory used by writes before starting to write on
> > > disk. When using tcpdump intensively, I lower this one to about 1%.
> > 
> > Hi Willy,
> > 
> > I know you're doing a great job on keeping the 2.4 kernel in shape,
> > but do you also have a good advice for people with more recent
> > kernels?  (hint: the file /proc/sys/vm/bdflush is missing)
> 
> OK OK OK... Next time I will have coffee *before* replying :-)
> 
> Check /proc/sys/vm/dirty_ratio and dirty_background_ratio. Both are
> percentage of total memory. The first one is for "foreground" writes
> (ie the writing process may block) while the second one is for
> "background" writes :
> 
> $ uname -a
> Linux hp 2.6.16-rc2-pa1 #1 Fri Feb 3 23:34:56 MST 2006 parisc unknown
> $ cat /proc/sys/vm/dirty_ratio 
> 40
> $ cat /proc/sys/vm/dirty_background_ratio 
> 10
> 
> Again, lowering those values should help writing data to disk
> sooner. Also you should take a look at min_free_kbytes (although
> I've not played with it yet) :

Ahh, okay, I didn't really understand these parameters before.
Now I think I understand what they are supposed to do.
I'll do some experiments with them.

Thanks for your help.
Toon.

BTW: That's pretty exotic hardware you have there (parisc).  ;-)
-
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/