The kernel slab cache allocator is doing something called colouring. It
works like this: When a slab is allocated and split up to the
structures, some space is left over. Thus the first structure can start
at offset from 0 to the amount of space left. The allocator thus chooses
all suitable offsets in turn (where suitable means aligned to specified
order with reasonable (for given platform) default).
Note, that this happens even for generic caches (kmalloc), where the
size is power of two. This is because for all caches with slab size
equal the page size the slab management data is at the end of the page,
so the space available is really PAGE_SIZE - 96.
I however don't know, if you are not talking about something more
sofisticated. (See comment in slab.c for reference to whitepaper about
the mechanizm implemented).
Thanks for the information, basically i am interested in knowing how linux kernel manages data allocation so that conflict misses in the data cache is minimized, the one you pointed out might also do the same, i am looking in there now.
when i google for "cache coloring linux SGI ", i am getting this
oss.sgi.com/projects/sgilinux-combined/ download/1.0-patches
but it seems this link is no more there, is it being integrated into the kernel, can i getit from some other place
Thanhs
Kugan
w iy!^Wvmdzޗ+r{mfX zޖwn'PmzY+
