Re: [PATCH] Scalable page cache

From: Andrew Morton (akpm@zip.com.au)
Date: Mon Nov 26 2001 - 14:29:19 EST


"David S. Miller" wrote:
>
> From: Ingo Molnar <mingo@elte.hu>
> Date: Mon, 26 Nov 2001 21:29:39 +0100 (CET)
>
> so i'm not against removing (or improving) the hash [our patch in fact
> just left the hash alone], but the patch presented is not a win IMO.
>
> Maybe you should give it a test to find out for sure :)

umm.. I've never seen any numbers from you, David.

Correct me if I'm wrong, but the pagecache_hash cost is
significant in the following situations:

1: TUX, because its pagecache lookups are not associated with
   a page copy. This copy makes the benefits of the patch
   unmeasurable with other workloads.

1a: Other sendfile-intensive applications. (Theoretical benefit.
    No benchmark results have been seen).

2: NUMA hardware, where the cost of cacheline transfer is much
   higher.

ergo, there is no point in futzing with the pagecache_lock *at all*
until either TUX is merged, or we decide to support large-scale
NUMA hardware well, which will require changes in other places.

Prove me wrong. Please.

-
-
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 : Fri Nov 30 2001 - 21:00:23 EST