2002-01-28 04:53:54

by Louis E Garcia II

[permalink] [raw]
Subject: Rik van Riel's vm-rmap


Does he still use classzones as the basis for the vm? I thought that
linux was trying to get away from classzones for better NUMA support in
2.5??

--Louis




2002-01-28 05:27:42

by William Lee Irwin III

[permalink] [raw]
Subject: Re: Rik van Riel's vm-rmap

On Sun, Jan 27, 2002 at 11:56:50PM -0500, Louis Garcia wrote:
> Does he still use classzones as the basis for the vm? I thought that
> linux was trying to get away from classzones for better NUMA support in
> 2.5??

rmap does not use the classzone concept.


Cheers,
Bill

2002-01-28 08:22:21

by Rik van Riel

[permalink] [raw]
Subject: Re: Rik van Riel's vm-rmap

On 27 Jan 2002, Louis Garcia wrote:

> Does he still use classzones as the basis for the vm? I thought that
> linux was trying to get away from classzones for better NUMA support in
> 2.5??

Nope. I've done a few modifications:

1) the IMHO inflexible classzone stuff has been removed

2) we have reverse mappings, so we can do our pageout
scan by physical address

3) this in turn means the active, inactive_dirty and
inactive_clean lists are per zone ... allowing us
to scan only in those zones where we actually need
to free pages

regards,

Rik
--
"Linux holds advantages over the single-vendor commercial OS"
-- Microsoft's "Competing with Linux" document

http://www.surriel.com/ http://distro.conectiva.com/

2002-01-29 00:04:07

by Louis E Garcia II

[permalink] [raw]
Subject: Re: Rik van Riel's vm-rmap

Does this patch work well with Andrew's low-latency patch?

--Louis


On Mon, 2002-01-28 at 03:21, Rik van Riel wrote:
> On 27 Jan 2002, Louis Garcia wrote:
>
> > Does he still use classzones as the basis for the vm? I thought that
> > linux was trying to get away from classzones for better NUMA support in
> > 2.5??
>
> Nope. I've done a few modifications:
>
> 1) the IMHO inflexible classzone stuff has been removed
>
> 2) we have reverse mappings, so we can do our pageout
> scan by physical address
>
> 3) this in turn means the active, inactive_dirty and
> inactive_clean lists are per zone ... allowing us
> to scan only in those zones where we actually need
> to free pages
>
> regards,
>
> Rik
> --
> "Linux holds advantages over the single-vendor commercial OS"
> -- Microsoft's "Competing with Linux" document
>
> http://www.surriel.com/ http://distro.conectiva.com/
>



2002-01-29 00:07:57

by Robert Love

[permalink] [raw]
Subject: Re: Rik van Riel's vm-rmap

On Mon, 2002-01-28 at 19:07, Louis Garcia wrote:
> Does this patch work well with Andrew's low-latency patch?

There will be failed hunks in the VM code (specifically vmscan.c), but
you can safely ignore them. So, yes, it works.

Robert Love

2002-01-29 00:11:17

by Louis E Garcia II

[permalink] [raw]
Subject: Re: Rik van Riel's vm-rmap

Should I do the rmap patch first?

--Louis

On Mon, 2002-01-28 at 19:12, Robert Love wrote:
> On Mon, 2002-01-28 at 19:07, Louis Garcia wrote:
> > Does this patch work well with Andrew's low-latency patch?
>
> There will be failed hunks in the VM code (specifically vmscan.c), but
> you can safely ignore them. So, yes, it works.
>
> Robert Love
>



2002-01-29 00:14:17

by Robert Love

[permalink] [raw]
Subject: Re: Rik van Riel's vm-rmap

On Mon, 2002-01-28 at 19:14, Louis Garcia wrote:
> Should I do the rmap patch first?

Yes, if Andrew's low-latency patch fails you just lose the scheduling
points in vmscan.c (big deal). If rmap half applies... ouch.

So apply rmap first.

Robert Love

2002-01-29 00:15:38

by Rik van Riel

[permalink] [raw]
Subject: Re: Rik van Riel's vm-rmap

On 28 Jan 2002, Louis Garcia wrote:

> Should I do the rmap patch first?

Yes.

After that you can patch the low latency patch,
which will give you a reject on vmscan.c

This doesn't matter because:
1) each part of the low latency patch is independant
2) -rmap already has low latency code in vmscan.c

kind regards,

Rik
--
"Linux holds advantages over the single-vendor commercial OS"
-- Microsoft's "Competing with Linux" document

http://www.surriel.com/ http://distro.conectiva.com/

2002-01-29 00:53:51

by Louis E Garcia II

[permalink] [raw]
Subject: Re: Rik van Riel's vm-rmap

Ok, has anyone ported the latest patch to 2.4.18-pre7 yet??

--Louis


On Mon, 2002-01-28 at 19:15, Rik van Riel wrote:
> On 28 Jan 2002, Louis Garcia wrote:
>
> > Should I do the rmap patch first?
>
> Yes.
>
> After that you can patch the low latency patch,
> which will give you a reject on vmscan.c
>
> This doesn't matter because:
> 1) each part of the low latency patch is independant
> 2) -rmap already has low latency code in vmscan.c
>
> kind regards,
>
> Rik
> --
> "Linux holds advantages over the single-vendor commercial OS"
> -- Microsoft's "Competing with Linux" document
>
> http://www.surriel.com/ http://distro.conectiva.com/
>



2002-01-29 01:39:57

by Alex Davis

[permalink] [raw]
Subject: Re: Rik van Riel's vm-rmap

Assuming you're talking about rmap12a, yes.
on http://www.dynamicbullet.com/rmap.html
there's a link to a patch that adds rmap
to 2.4.18pre7


-Alex

>Ok, has anyone ported the latest patch to 2.4.18-pre7 yet??
>
>--Louis




__________________________________________________
Do You Yahoo!?
Send your FREE holiday greetings online!
http://greetings.yahoo.com