Re: Major regression on hackbench with SLUB (more numbers)

[Ingo Molnar <mingo@elte.hu>]

* Ingo Molnar <mingo@elte.hu> wrote:

> I'd also not rely on the fact that only a few people are complaining. 
> Most people, even 2.6.24-rc early adopters, still use SLAB because 
> early adopters typically use their .23 .config and do a 'make 
> oldconfig' - which picks up SLAB. So SLUB use will become more 
> widespread only once 2.6.24 is out and is packaged in distros. Distros 
> will likely pick SLUB if there's no performance worries and if it's 
> the default. Fedora rawhide already uses SLUB.

here's some silly statistics about allocator coverage on lkml, based on 
configs reported to lkml in the past 4 months:

  $ for N in SLAB SLUB SLOB; do printf "%s: " $N; grep ^CONFIG_$N=y linux-kernel | wc -l; done
  SLAB: 70
  SLUB: 77
  SLOB: 4

so SLUB and SLAB is utilized about equally amongst people who reported 
configs to lkml.

But people who use SLUB enabled it intentionally - and they are thus 
much less likely to complain about this choice of them.

Reporting:

 "I just enabled SLUB instead of SLAB, and hey guys, it does not have 
  SLABinfo"

has a foolish ring to it, doesnt it? I'd rather ask carefully, like this 
person did:

  http://kerneltrap.org/mailarchive/linux-kernel/2007/10/12/335765

This is one of the reasons why i think the whole SLAB->SLUB renaming was 
bad - we should have done SLAB2 or SLAB^2 instead, so that people expect 
_at least as good_ behavior (performance, features, etc.) from SLUB as 
from SLAB. Instead of "something different".

anyway ... i think we still generally suck _alot_ at providing 
near-transparent kernel upgrades to users. (kernel upgrades are still a 
pain and risk, and often just due to poor developer choices on our 
side.)

It's nowhere near as bad as the 2.4->2.6 transition was (in fact it 
shouldnt even be mentioned in the same sentence), and it's getting 
better gradually, but i think we should just by default be 10 times more 
careful about these things - whenever it is borderline technically sane 
to do so. We induce enough unintentional breakage of user-space, we 
shouldnt compound it with intentional breakages. The kernel community 
still has _a lot_ of user and distro trust to win back. So being seen as 
over-cautious wont harm.

	Ingo