Re: [PATCH 2/2] Call percpu smp cacheline algin interface

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 9 May 2007 16:10:05 -0700
"Yu, Fenghua" <fenghua.yu@intel.com> wrote:

> 
> 
> On Wed, 9 May 2007, Andrew Morton wrote:
> 
> >> erm, it's not obviosu from all this that the patches are worth
> proceeding
> >> with, are they?
> 
> >What was it? 0.5% performance improvement on a synthetic benchmark? 
> >Process wakeup I believe?
> 
> The initial patch and discussion is from:
> http://www.uwsg.iu.edu/hypermail/linux/kernel/0704.1/0340.html
> 
> Yes, the runqueue patch has a 0.5% perf improvement on database
> workload(which is a good improvement for this workload).
> 
> The theory behind the patches is:
> 
> 1. Minimize number of cache lines that are touched during a remote
> access. On Numa system, remote access is more expensive than local.
> 2. Do not share cache line between remote accessed data and local
> accessed data. Local data update may cause remote access cache miss and
> wait for longer time.
> 
> Although the patches themselves don't save or waste per_cpu size, the
> above two reasons are good to have them in.
> 

Guys, this is all a lesson in the value of changelogs, and in how not to
write them.

Can you please prepare a new changelog for these patches?  Something which
encapsulates all the above in as brief a form as possible and which
includes some numbers describing the space and/or speed improvements?

Thanks.