On Mon, May 6, 2019 at 10:57 PM Johannes Weiner wrote:
>
> On Sun, Apr 28, 2019 at 03:44:34PM +0800, Zhaoyang Huang wrote:
> > From: Zhaoyang Huang
> >
> > this patch introduce timestamp into workingset's entry and judge if the
> > page is
> > active or inactive via active_file/refault_ratio ins
On Sun, Apr 28, 2019 at 03:44:34PM +0800, Zhaoyang Huang wrote:
> From: Zhaoyang Huang
>
> this patch introduce timestamp into workingset's entry and judge if the page
> is
> active or inactive via active_file/refault_ratio instead of refault distance.
>
> The original thought is coming from th
From: Zhaoyang Huang
this patch introduce timestamp into workingset's entry and judge if the page is
active or inactive via active_file/refault_ratio instead of refault distance.
The original thought is coming from the logs we got from trace_printk in this
patch, we can find about 1/5 of the fil
rebase the commit to latest mainline and update the code.
@Matthew, with regarding to your comment, I would like to say the
algorithm doesn't change at all. I do NOT judge the page's activity
via an absolute time value, but still the refault distance. What I
want to fix is the scenario which drop l
On Wed, Apr 17, 2019 at 08:26:22PM +0800, Zhaoyang Huang wrote:
[quoting Johannes here]
> As Matthew says, you are fairly randomly making refault activations
> more aggressive (especially with that timestamp unpacking bug), and
> while that expectedly boosts workload transition / startup, it comes
On Wed 17-04-19 20:26:22, Zhaoyang Huang wrote:
> repost the feedback by under Johannes's comment
Please follow up in the original email thread. Fragmenting the
discussion is exactly what I wanted...
--
Michal Hocko
SUSE Labs
repost the feedback by under Johannes's comment
When something like a higher-order allocation drops a large number of
file pages, it's *intentional* that the pages that were evicted before
them become less valuable and less likely to be activated on refault.
There is a finite amount of in-memory LR
On Wed 17-04-19 19:36:21, Zhaoyang Huang wrote:
> sorry for the confusion. What I mean is the basic idea doesn't change
> as replacing the refault criteria from refault_distance to timestamp.
> But the detailed implementation changed a lot, including fix bugs,
> update the way of packing the timest
sorry for the confusion. What I mean is the basic idea doesn't change
as replacing the refault criteria from refault_distance to timestamp.
But the detailed implementation changed a lot, including fix bugs,
update the way of packing the timestamp, 32bit/64bit differentiation
etc. So it makes sense
On Wed 17-04-19 18:55:15, Zhaoyang Huang wrote:
> fix one mailbox and update for some information
>
> Comparing to
> http://lkml.kernel.org/r/1554348617-12897-1-git-send-email-huangzhaoy...@gmail.com,
> this commit fix the packing order error and add trace_printk for
> reference debug information
fix one mailbox and update for some information
Comparing to
http://lkml.kernel.org/r/1554348617-12897-1-git-send-email-huangzhaoy...@gmail.com,
this commit fix the packing order error and add trace_printk for
reference debug information.
For johannes's comments, please find bellowing for my fee
Hi,
I do not see
http://lkml.kernel.org/r/1554348617-12897-1-git-send-email-huangzhaoy...@gmail.com
discussion reaching a conlusion to change the current workingset
implementation. Therefore is there any reason to post a new version of
the patch? If yes it would be really great to see a short summ
add Johannes and answer his previous question.
@Johannes Weiner
Yes. I do agree with you about the original thought of sacrificing
long distance access pages when huge memory demands arise. The problem
is what is the criteria of the distance, which you can find from what
I comment in the patch, th
From: Zhaoyang Huang
This patch introduce timestamp into workingset's entry and judge if the page
is active or inactive via active_file/refault_ratio instead of refault distance.
The original thought is coming from the logs we got from trace_printk in this
patch, we can find about 1/5 of the fil
14 matches
Mail list logo