Tigran Aivazian wrote:
>
> HI Manfred,
>
> You changed some __get_free_page(GFP_KERNEL) to kmalloc(PAGE_SIZE,
> GFP_KERNEL).
>
Yes, this has a huge effect:
gfp is _very_ slow, kmalloc is around 600 cpu ticks faster, from a total
of 2000 cpu ticks for stat("/bin").
> Even if so, switching to kmalloc makes it possible to fail more often (or
> sooner) than plain gfp - have you considered this?
I'll check that, but I think that kmalloc(PAGE_SIZE) will internally
call gfp(order==1).
But the patches are on-hold:
* Ingo has written a per-cpu slab. I'll wait until I have seen his
patch.
* gfp could be speed-up: IMHO we should switch from the normal buddy to
a lazy buddy: as long as noone asks for order>0, we can avoid coalescing
all pages. I found a reference in "Unix Internals" [the book mentioned
in mm/slab.c], I'll check that today.
-- Manfred- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed May 31 2000 - 21:00:11 EST