Re: [PATCH] 2/2 Prezeroing large blocks of pages during allocationVersion 4

[Mel Gorman]

On Tue, 8 Mar 2005, KAMEZAWA Hiroyuki wrote:

> Mel Gorman wrote:
>
> > > >
> > > Now, 5bits per  MAX_ORDER pages.
> > > I think it is simpler to use "char[]" for representing type of  memory
> > > alloc
> > > type than bitmap.
> > >
> > >
> >
> > Possibly, but it would also use up that bit more space. That map could be
> > condensed to 3 bits but would make it that bit (no pun) more complex and
> > difficult to merge. On the other hand, it would be faster to use a char[]
> > as it would be an array-index lookup to get a pageblock type rather than a
> > number of bit operations.
> >
> > So, it depends on what people know to be better in general because I have
> > not measured it to know for a fact. Is it better to use char[] and use
> > array indexes rather than bit operations or is it better to leave it as a
> > bitmap and condense it later when things have settled down?
> >
> Hmm, Okay, I'll wait for condensed version.
> BTW, in space consumption/cache view,  does using bitmap have  real benefit
> ?
>

For space, there is a small benefit. On my system with 1.5GiB of RAM, it
is about 130 bytes saved for prezeroing and about 220 with just the
placement policy.  For speed, I do not know how bitmaps normally perform
with the CPU cache, but for the placement policy, it makes no difference.
I implemented a version using char[] array and there was no performance
difference that I could measure. The bitmaps just are not consulted often
enough to make a big performance difference.

-- 
Mel Gorman
Part-time Phd Student				Java Applications Developer
University of Limerick				    IBM Dublin Software Lab
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/