> 2.1.131 is out there now - and will be the last kernel release for a
> while. I'm going to Finland for a week and a half, and will be back
> mid December. During that time I hope people will beat on this.
There are a few things in the VM balancing of 2.1.131 that
I strongly disagree with and that will make performance
far worse than it could have been.
+++ linux/mm/swap.c Fri Nov 27 10:15:29 1998
- 25, /* borrow percent buffer */
+ 10, /* borrow percent buffer */
- 30, /* borrow percent page cache */
+ 15, /* borrow percent page cache */
Decreasing the borrow percentages will make the system
clear out the caches too soon. The system should basically
be self-tuning and the borrow percentages are only there
to make sure that the system doesn't continue swapping
when we have obscene amounts of cache.
+++ linux/mm/vmscan.c Mon Nov 30 16:05:40 1998
@@ -431,7 +431,7 @@
kmem_cache_reap(gfp_mask);
if (buffer_over_borrow() || pgcache_over_borrow())
- shrink_mmap(i, gfp_mask);
+ state = 0;
This is a good fix as it returns behaviour back to the normal
and desired behaviour (except for too heavy cache pruning).
This patch would have been better however, since it does what
we want without too heavy cache pruning or massive swapout
frenzies.
+++ ./mm/vmscan.c Tue Dec 1 07:12:28 1998
@@ -431,6 +431,8 @@
kmem_cache_reap(gfp_mask);
if (buffer_over_borrow() || pgcache_over_borrow())
+ state = 0;
+ if (atomic_read(&nr_async_pages) > pager_daemon.swap_cluster / 2)
shrink_mmap(i, gfp_mask);
switch (state) {
This patch (against 2.1.130) makes sure that kswapd will free
memory even during heavy swapout frenzies. This makes sure
that kswapd will finish (swapout frenzies don't free memory
and make kswapd swap too much without doing it's job) without
swapping too heavily.
By testing for swap_cluster/2 we make sure that we don't hamper
swapout throughput too much and don't block the I/O subsystem
either.
I have been using this patch for several days now without any
problems whatsoever (and far better performance than ever before).
And finally here is a third patchlet to fix page_io.c, it should
be fairly obvious what this patch does. It is mainly useful when
we have multiple swap partitions over which we stripe -- increasing
the value of swap_cluster will give us much more throughput then.
--- ./mm/page_io.c.orig Thu Nov 26 11:26:49 1998
+++ ./mm/page_io.c Thu Nov 26 11:30:43 1998
@@ -60,7 +60,7 @@
}
/* Don't allow too many pending pages in flight.. */
- if (atomic_read(&nr_async_pages) > SWAP_CLUSTER_MAX)
+ if (atomic_read(&nr_async_pages) > pager_daemon.swap_cluster)
wait = 1;
p = &swap_info[type];
regards,
Rik -- the flu hits, the flu hits, the flu hits -- MORE
+-------------------------------------------------------------------+
| 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
Please read the FAQ at http://www.tux.org/lkml/