> I've just switched from 2.1.129 to 2.1.131 + Rik's "fastest VM system"
> patch, and I have to say, nice!
>
> Disk activity
> =============
>
> Something's changed. Can't say whether it's the main tree changes or
> Rik's mm changes (I haven't tried plain 2.1.131).
Both. Stephen's changes to shrink_mmap() really have a huge
influence on performance.
> I'm not using any swap either (well, 16k on a 64MB system). It feels
> faster somehow anyway.
The not-using-any-swap part comes both from Stephen's improvement
and my change to vmscan.c; we now quit a swap_out() loop earlier
and we do (over?-)agressive pruning of the page and buffer caches.
> Well, apart from Squid which still spends a few minutes grinding
> away at the disk when it starts. I would like to find a way to fix
> this, it's probably the most thrashy thing my disk has to handle and
> it does slow down everything else for a while quite significantly.
This could be a side effect of a too agressive pruning of the
caches. We should fix this.
> Netscape still hits the disk very hard when it starts, and takes what
> seems like just as long. Netscape is pretty quick to start the second
> time though (about 3 seconds), so it's definitely a paging thing. Is
> there anything which could be done with paging
> read-ahead/read-behind/read-cleverer to make Netscape not thrash the
> disk when it starts?
No, not really. When Netscape starts it is not in cache yet :)
I agree that we should do better readbehind though. I am
currently designing a nice algorithm for deciding whether
to do read-ahead, read-behind, a bit of both or a lot of
both. It is mainly focused at swap I/O though...
File I/O might be better served with a slightly different
algorithm, so we need a volunteer for that... If you are
that volunteer: linux-mm@kvack.org is the list to be.
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/