Re: [RFC] 2.3.39 zone balancing

[Kanoj Sarcar]

> 
> Hi,
> 
> On Tue, 8 Feb 2000 12:52:51 +0100 (CET), Andrea Arcangeli
> <[e-mail]> said:
> 
> > I think freepages.* is dead. I just wanted to remove it but actually
> > nobody is hurted by it so I left it there for now.
> 
> freepages.* is critically important for allowing the user to define the
> "headroom" of free memory maintained for ATOMIC allocations.  On busy
> networked machines, this is absolutely vital.  For user-mode
> allocations, it's never going to be that important except as a way of
> allowing some slack as we start up the page stealer in the background.
> 
> --Stephen
> 

Freepages.* is a global counter, that's why I think we should have a
per zone freepages.*. For 2.3, atomic highmem allocations are not an
issue (almost), so that leaves atomic dma and normal requests. On a
machine with a lot of free normal memory, the number of free dma pages 
might be running low, but that will not trigger kswapd, since 
nr_free_pages() will give an aggregrate sum, which will seem fine
to kswapd when compared to freepages.*.

Okay, if that is believable logic for a zone-aware kswapd (and per zone
freepages.*), the per zone freepages.* can be rolled into how the zone 
pages_low/pages_high etc values are setup when the zone is initialized. 

Kanoj