Re: cp file /dev/zero <-> cache [was Re: increasing page size]

Rik van Riel (H.H.vanRiel@phys.uu.nl)
Wed, 8 Jul 1998 21:03:50 +0200 (CEST)

On Wed, 8 Jul 1998, Stephen C. Tweedie wrote:
> <H.H.vanRiel@phys.uu.nl> said:
>
> > When a file consumes more than a certain percentage
> > of the page cache and the page cache is larger than
> > a certain percentage of memory, we must grab a page
> > from the page cache instead of getting a new page...
>
> There's a better way (I've already got this on the way for 2.1, and will
> be experimenting with the performance to see whether it has unexpected
> nasty side-effects): don't _force_ a cache reclaim, but simply set the
> cache as the preferential source in try_to_free_page(). It's the most

It's been in the kernel since 2.1.90, or thereabouts...
[vmscan.c:do_try_to_free_page()]
Andrea has clearly pointed out that it doesn't work on small
memory machines.

> There are potential problems with this behaviour, but it "feels" right:
> it is inherently self-balancing and has no arbitrary rules to govern the
> cache size at any moment in time.

It is self-balancing, just not enough so. Besides, my rule
didn't place a limit on the cache size, it _only_ forces
a cache reclamation when:
- the cache is > xx% of memory AND
- the inode in question takes > xx% of the cache

This means that the cache won't be trimmed when we
need to cache some innocent small file.

Rik.
+-------------------------------------------------------------------+
| Linux memory management tour guide. H.H.vanRiel@phys.uu.nl |
| Scouting Vries cubscout leader. http://www.phys.uu.nl/~riel/ |
+-------------------------------------------------------------------+

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu