Re: 2.4.10 does not set accessed bit for readahead pages

From: Linus Torvalds (torvalds@transmeta.com)
Date: Fri Sep 28 2001 - 16:58:07 EST


On Fri, 28 Sep 2001, Marcelo Tosatti wrote:
>
> I've done some experiments in the past trying to use a new GFP_ flag for
> allocation of readahead pages. This new GFP flag was the indicator to
> __alloc_pages() that it should fail very easily (so, in theory, we would
> not spend time on page reclaiming for readahead, which is simply a guess).

I don't think the above has anything to do with the accessed bit.

The fact is, failing very easily will mean that we never do much
read-ahead at all, or will mean that _other_ allocations will have to bear
more of the grunt of the work, and that the LRU lists just become
unbalanced. I doubt that the fact that your GFP experiments back up
touching the accessed bit.

However, if somebody has benchmarks comparing apples to apples (ie
comparing _just_ setting/notsetting the accessed bit), that would be very
interesting.

Not setting the accessed bit means that the read-ahead has to be useful
within "one loop" of the inactive list. I think that's the right thing: if
we don't actually touch the pages soon, the read-ahead was probably a
loss. And it's better to cut your losses early rather than late.

                Linus

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Sun Sep 30 2001 - 21:01:05 EST