Re: writeback list

Re: writeback list

[Rajagopal Ananthanarayanan]

Rik van Riel wrote:
> 
> Hi Stephen,
> 
> we may have forgotten something in our new new vm design from
> last weekend. While we have the list head available to put
> pages in the writeback list, we don't have an entry in to put
> the timestamp of the write in struct_page...
> 
> Maybe we want to have an active list after all and replace the
> buffer_head pointer with a pointer to another structure that
> tracks the writeback stuff that's now tracked by the buffer head?
> 
> (things like: prev, next, write_time and a few other things)
> 


Yes, maintaining time information in the page will be useful
for XFS also. Basically, there are pages in the page cache
without a particular block(s) assigned to the page ... these
are the delayed allocate pages. Such pages don't have any
buffer_heads associated with them, until the delalloc is converted.

It will be great if the delalloc pages can be somehow temporally ordered.
The write-back list you propose seems to fit the bill nicely.

regards,

ananth.


-- 
--------------------------------------------------------------------------
Rajagopal Ananthanarayanan ("ananth")
Member Technical Staff, SGI.
--------------------------------------------------------------------------
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: writeback list

[Rik van Riel]

On Thu, 13 Jul 2000, Rajagopal Ananthanarayanan wrote:
> Rik van Riel wrote:

> > we may have forgotten something in our new new vm design from
> > last weekend. While we have the list head available to put
> > pages in the writeback list, we don't have an entry in to put
> > the timestamp of the write in struct_page...
> > 
> > Maybe we want to have an active list after all and replace the
> > buffer_head pointer with a pointer to another structure that
> > tracks the writeback stuff that's now tracked by the buffer head?
> > 
> > (things like: prev, next, write_time and a few other things)
> 
> Yes, maintaining time information in the page will be useful for
> XFS also. Basically, there are pages in the page cache without a
> particular block(s) assigned to the page ... these are the
> delayed allocate pages. Such pages don't have any buffer_heads
> associated with them, until the delalloc is converted.

Exactly, this is what the write-back list will be used for.

> It will be great if the delalloc pages can be somehow temporally
> ordered. The write-back list you propose seems to fit the bill
> nicely.

To put it more strongly, the write-back list *needs* to be
temporally ordered if we want to have a kupdate like we have
today.

regards,

Rik
--
"What you're running that piece of shit Gnome?!?!"
       -- Miguel de Icaza, UKUUG 2000

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

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

writeback list

[Rik van Riel]

Hi Stephen,

we may have forgotten something in our new new vm design from
last weekend. While we have the list head available to put
pages in the writeback list, we don't have an entry in to put
the timestamp of the write in struct_page...

Maybe we want to have an active list after all and replace the
buffer_head pointer with a pointer to another structure that
tracks the writeback stuff that's now tracked by the buffer head?

(things like: prev, next, write_time and a few other things)

regards,

Rik
--
"What you're running that piece of shit Gnome?!?!"
       -- Miguel de Icaza, UKUUG 2000

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

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: writeback list

[Stephen C. Tweedie]

Hi,

On Thu, Jul 13, 2000 at 04:30:35PM -0300, Rik van Riel wrote:
> 
> we may have forgotten something in our new new vm design from
> last weekend. While we have the list head available to put
> pages in the writeback list, we don't have an entry in to put
> the timestamp of the write in struct_page...

It shouldn't matter.  Just assume something like the 30-second sync.
You can keep placeholders in the list for that, or even do something
like have multiple lists, one for the current 30-seconds being synced,
one for the next sync.  You can do the same for 5-second metadata
syncs too if you want; just use separate lists.  It perturbs the LRU a
bit, but then we also have page aging so that's not too bad.  (Are you
planning on using the age values in the inactive list, btw?)

Cheers,
 Stephen
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/

Re: writeback list

[Rik van Riel]

On Fri, 14 Jul 2000, Stephen C. Tweedie wrote:
> On Thu, Jul 13, 2000 at 04:30:35PM -0300, Rik van Riel wrote:
> > 
> > we may have forgotten something in our new new vm design from
> > last weekend. While we have the list head available to put
> > pages in the writeback list, we don't have an entry in to put
> > the timestamp of the write in struct_page...
> 
> It shouldn't matter.  Just assume something like the 30-second sync.
> You can keep placeholders in the list for that, or even do something
> like have multiple lists, one for the current 30-seconds being synced,
> one for the next sync.  You can do the same for 5-second metadata
> syncs too if you want;

Placeholders seem like the best idea to me. Having separate
lists for data and metadata will have to go away anyway when
we start using journaled filesystems (which have their own
ordering issues).

> It perturbs the LRU a bit, but then we also have page aging so
> that's not too bad.  (Are you planning on using the age values
> in the inactive list, btw?)

All pages on the inactive list will have age 0, once they are
touched they go back to the active list...

regards,

Rik
--
"What you're running that piece of shit Gnome?!?!"
       -- Miguel de Icaza, UKUUG 2000

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

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux.eu.org/Linux-MM/