FYI.
I've implemented Solarias's "Slab Allocator" for Linux.
This is a fast allocator for common objects, and espically those that
have equivalent states at construct and desconstruct time. It also
tries to be hardware cache friendly (small footprint, cache colouring).
As it had no chance of getting into 2.0, I haven't finished all those
little 'jobs' off (eg. debugging support, such as RedZoning). Hope to
get this done by the end of next week.
I've also implemented SVR4's "Power-of-2 Lazy Buddy Allocator", for
general purpose memory allocations. It's in user-space at the
moment, and in need of much tidying-up. This should be faster than
the current 'kmalloc' for general purpose allocations, but suffers
two major draw-backs;
o For efficient freeing, the size needs to be passed in
o Freeing performs only the most basic sanity checks on the
given address.
The second point will probably kill the 'Lazy-Buddy' off for Linux...
markhe