> The whole point of the cache is to decrease disk-reads....the
> current implementation does not take into account the fact that swapping
> implies disk reads! I suspect that this problem is only exposed to the
> light of day because the pte-chaining/defragmentation stuff isn't in the
> kernel yet. But it is STILL a problem! Is there a reason why we should
> not implement at least a default mechanism for figuring out when it is
> better to uncache or to swap? That default mechanism could then be
> improved with time.
It's there. You can even tune it through /proc/sys/vm/*
The change we'll probably need is changing the swap policy
to SC_ROUNDROBIN from SC_whatwehavenow.
(see the 1.2.xx source code for references :)
I'll work on it ASAP
Rik.
+-------------------------------------------+--------------------------+
| Linux: - LinuxHQ MM-patches page | Scouting webmaster |
| - kswapd ask-him & complain-to guy | Vries cubscout leader |
| http://www.phys.uu.nl/~riel/ | <H.H.vanRiel@phys.uu.nl> |
+-------------------------------------------+--------------------------+
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu