Re: [PATCH 0/2] pdflush fix and enhancement

[Martin Knoblauch <spamtrap@knobisoft.de>]

please CC me on replies, I am not subscribed to LKML.


----- Original Message ----
> From: Andi Kleen <andi@firstfloor.org>
> To: Peter W. Morreale <pmorreale@novell.com>
> Cc: Andi Kleen <andi@firstfloor.org>; linux-kernel@vger.kernel.org
> Sent: Wednesday, December 31, 2008 3:46:09 AM
> Subject: Re: [PATCH 0/2] pdflush fix and enhancement
> 

snip

> > I actually think the question is: Why not allow the admin to control
> > this?  Since it seems like this is a matter of policy based on machine
> > configuration. 
> 
> The kernel should know the current machine config and most 
> admins don't really want to do very fine grained configuration;
> they expect the system to perform well out of the box. That is
> why it is adventageous to try to come up with good auto tuning.
> 

 Independent of the patch in question, the problem with this seems to me that [some/many of] the kernel developers [seem to] try to get it right for 100% of all thinkable use-cases. But this fails to take into account that:

- you cannot think of every single use-case. And not only because predicting furure use-cases is difficult
- getting it right for every case very often creates complexity that leads to subtle problems tha are hard to analyse and fix
- it may waste developer ressources
- and, think about it,  do we really want  the kernel to be smarter than ourselves ? :-)

 You are right that the kernel should work out of the box most of the times. And it usually is pretty good at that. But there are corner-cases where more flexibility for the admins is desirable - if only to debug a problem without doing deep code hacking. So we should be careful adding tuning-knobs, but we should also admit that sometimes they are useful.


Happy New Year
Martin