* dm-bufio cache patches
@ 2014-10-08 20:10 Mikulas Patocka
2014-10-08 20:17 ` Mikulas Patocka
0 siblings, 1 reply; 5+ messages in thread
From: Mikulas Patocka @ 2014-10-08 20:10 UTC (permalink / raw)
To: Edward Thornber, Mike Snitzer; +Cc: dm-devel, Alasdair G. Kergon
Hi
I created dm-bufio patchset that keep some amount of cache indefinitely,
regardless of the timeout (the default is to keep 1/500 of total memory
indefinitely).
I uploaded it here:
http://people.redhat.com/~mpatocka/patches/kernel/dm-bufio-cache/
Try it, if it helps, we can backport it.
If you think that we should use different default values (for example,
increase the max_age value), say it.
Mikulas
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: dm-bufio cache patches
2014-10-08 20:10 dm-bufio cache patches Mikulas Patocka
@ 2014-10-08 20:17 ` Mikulas Patocka
2014-10-09 13:07 ` Joe Thornber
0 siblings, 1 reply; 5+ messages in thread
From: Mikulas Patocka @ 2014-10-08 20:17 UTC (permalink / raw)
To: Edward Thornber, Mike Snitzer; +Cc: dm-devel, Alasdair G. Kergon
On Wed, 8 Oct 2014, Mikulas Patocka wrote:
> Hi
>
> I created dm-bufio patchset that keep some amount of cache indefinitely,
> regardless of the timeout (the default is to keep 1/500 of total memory
> indefinitely).
>
> I uploaded it here:
> http://people.redhat.com/~mpatocka/patches/kernel/dm-bufio-cache/
>
> Try it, if it helps, we can backport it.
>
> If you think that we should use different default values (for example,
> increase the max_age value), say it.
BTW. I have no idea if the default values are good or not - so it would be
better if you come up with default values based on your benchmarks.
Mikulas
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: dm-bufio cache patches
2014-10-08 20:17 ` Mikulas Patocka
@ 2014-10-09 13:07 ` Joe Thornber
2014-10-09 15:40 ` Mikulas Patocka
0 siblings, 1 reply; 5+ messages in thread
From: Joe Thornber @ 2014-10-09 13:07 UTC (permalink / raw)
To: Mikulas Patocka; +Cc: Mike Snitzer, dm-devel, Alasdair G. Kergon
On Wed, Oct 08, 2014 at 04:17:59PM -0400, Mikulas Patocka wrote:
> > I created dm-bufio patchset that keep some amount of cache indefinitely,
> > regardless of the timeout (the default is to keep 1/500 of total memory
> > indefinitely).
I'm running something similar here. I just put in a hard coded
default rather than basing it on the amount of memory present. I'm
not sure which I prefer.
I've also switched over to using a rbtree instead of that huge hash
table which consumes 8m.
I'll post clean patches in a bit.
- Joe
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: dm-bufio cache patches
2014-10-09 13:07 ` Joe Thornber
@ 2014-10-09 15:40 ` Mikulas Patocka
2014-10-13 9:02 ` Joe Thornber
0 siblings, 1 reply; 5+ messages in thread
From: Mikulas Patocka @ 2014-10-09 15:40 UTC (permalink / raw)
To: Joe Thornber; +Cc: Mike Snitzer, dm-devel, Alasdair G. Kergon
On Thu, 9 Oct 2014, Joe Thornber wrote:
> On Wed, Oct 08, 2014 at 04:17:59PM -0400, Mikulas Patocka wrote:
> > > I created dm-bufio patchset that keep some amount of cache indefinitely,
> > > regardless of the timeout (the default is to keep 1/500 of total memory
> > > indefinitely).
>
> I'm running something similar here. I just put in a hard coded
> default rather than basing it on the amount of memory present. I'm
> not sure which I prefer.
>
> I've also switched over to using a rbtree instead of that huge hash
> table which consumes 8m.
>
> I'll post clean patches in a bit.
>
> - Joe
Does rbtree have better performance than the hash table?
Mikulas
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2014-10-13 9:02 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2014-10-08 20:10 dm-bufio cache patches Mikulas Patocka
2014-10-08 20:17 ` Mikulas Patocka
2014-10-09 13:07 ` Joe Thornber
2014-10-09 15:40 ` Mikulas Patocka
2014-10-13 9:02 ` Joe Thornber
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.