Subject: Re: swapctl -a doesn't do load-balancing; whyzzat?
To: Jason R Thorpe , Robert Elz <kre@munnari.OZ.AU>
From: Chuck Silvers <email@example.com>
Date: 09/03/2000 18:19:55
On Sun, Sep 03, 2000 at 03:37:31PM -0700, Jason R Thorpe wrote:
> On Mon, Sep 04, 2000 at 09:23:29AM +1100, Robert Elz wrote:
> > The only place I had excess space on the system concerned was via NFS, so
> > I made myself a nice huge swap file out there, and enabled swapping.
> > Almost before I could think, the system (which was paging its little head off
> > of course - that's why everything was taking so long in the first place)
> > had started moving all kinds of stuff from the local disc out to NFS
> > swapping, which didn't improve its performance of course... (And yes, I
> > had managed to forget to set the swap priority, so the NFS swap space
> > had defaulted to being equivalent to the local disc).
> The "migration" effect is probably the result of things being brought
> back in, the objects being deleted, and new ones being created, and
> thus pushed back out.
> As far as I know, once an object/offset or amap/offset has a swap slot
> assigned to it, the swap slot persists until the swap device is invalidated
> via swapctl -d.
the other way that the swapslot for a page can change is if the page is
modified. if the page is paged out again, the swapslot is reassigned.