2.6 kernel tuning guides

[Lennart Sorensen]

On Sun, Jul 24, 2005 at 09:53:03AM -0400, Fraser Campbell wrote:
> That much I know.  I was hoping to find detailed advice, as examples:
> 
> - precisely what is impact of changing /proc/sys/vm/dirty_*,
> - aio-max-size, what is it's equivalent in 2.6, is it a fixed size?
> - impact of different I/O schedulers according to type of application,
>   which scheduler is best where, what tunables are worth playing with?
> - impact of different I/O schedulers in UP versus SMP
> 
> Pieces of this are documented in kernel source but much of it would be
> difficult or impossible to deduce from source reading.  If Google is
> having a good day some stuff is even available online.  I had hoped to
> short circuit testing by having some good up-front guidelines.

Some of it is documented in the Documentation subdir of the kernel
source.  Some isn't really documented at all, and some is documented at
the top of the source code implementing it.  Pretty sad state to be in I
guess, but on the other hand, it's not like Microsoft documents all the
tuneable registry settings either.

> I did find a few books in Chapters (Square One) and early reading is
> promising:
> 
> - Performance Tuning for Linux Servers (IBM Press)
> - Optimizing Linux Performance (HP)
> 
> The HP book is focused a lot on monitoring tools (vmstat, iostat,
> mpstat, sar, ps, top, cachegrind, etc.) and how to pinpoint application
> bottlenecks.  The IBM book gets more into kernel specific tuning
> options.  Both books appear to have decent case studies.

I wonder if either applies to 2.6 or only to older kernels?

Lennart Sorensen
--
The Toronto Linux Users Group.      Meetings: http://tlug.ss.org
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://tlug.ss.org/subscribe.shtml