Re: [RFC][PATCH 0/3] Reduce number of avtab nodes

[Luke Kenneth Casson Leighton <lkcl@lkcl.net>]

On Mon, Aug 01, 2005 at 09:41:05AM -0400, Stephen Smalley wrote:
> On Sun, 2005-07-31 at 11:59 -0400, James Morris wrote:
> > With these patches, on a 64-bit system, after boot:
> > 
> >   
> >             #objs  objsize   kernmem
> > Targeted:
> >   Before:  237888       40     9.1MB
> >   After:    19968       24     468KB
> > 
> > Strict:
> >   Before:  571680       40   21.81MB
> >   After:   221052       24    5.06MB
> 
> Ok, this along with the data I posted for 32-bit is certainly
> encouraging as far as memory savings are concerned.  Things that we need
> to resolve before committing/upstreaming these changes include:
> 
> - We need to measure the performance impact of the changes on the
> kernel, if any, and possibly optimize the compute_av code if it is
> measurable.

 little story for you that is comp-sci related [which if
 you're aware of these sorts of things i humbly apologise for
 mentioning - grandmas, eggs, suck, sorry :) ]

 andrew tridgell wrote an upper-case / lower-case unicode
 converter for samba.  it wasn't a 16-bit lookup table,
 because 65536 * 2 equals 128kbytes which is _way_ more memory
 than can fit into a first level cache of some processors.
 consequently, performance of a lookup table would be crap.
 what he wrote instead was a small algorithm with lookup ranges.
 it was heaps faster.
 
 anyway, what you're doing is reminiscent of that and so
 i thought i should mention it and i invite you to consider
 performing some tests on hardware ranging from ARM 2-400mhz
 with the usual shit-for-brains caches through PIIIs and
 AMD K7s to P4s, picking the most likely candidates where
 it's going to matter or even possibly consider offering /
 enabling / disabling different optimisations depending on
 the processor target (CONFIG_xxx).

 again though i respectfully ask you to totally ignore me
 should you have considered these things already - comp-sci
 being what it is.

 l.


--
This message was distributed to subscribers of the selinux mailing list.
If you no longer wish to subscribe, send mail to majordomo@tycho.nsa.gov with
the words "unsubscribe selinux" without quotes as the message.