From: Luigi Semenzato <semenzato@google.com>
To: Minchan Kim <minchan@kernel.org>
Cc: Linux Memory Management List <linux-mm@kvack.org>,
Douglas Anderson <dianders@google.com>,
Dmitry Torokhov <dtor@google.com>,
Sonny Rao <sonnyrao@google.com>
Subject: Re: [PATCH] mm: add counters for different page fault types
Date: Fri, 26 May 2017 11:43:48 -0700 [thread overview]
Message-ID: <CAA25o9QG=Juynu-8wAYvdY1t7YNGVtE10fav2u3S-DikuU=aMQ@mail.gmail.com> (raw)
In-Reply-To: <20170526040622.GB17837@bbox>
Many thanks Minchan.
On Thu, May 25, 2017 at 9:06 PM, Minchan Kim <minchan@kernel.org> wrote:
> If it is swap cache hit, it's not a major fault which causes IO
> so VM count it as minor fault, not major.
Cool---but see below.
> Yub, I expected you guys used zram with readahead off so it shouldn't
> be a big problem.
By the way, I was referring to page clustering. We do this in sysctl.conf:
# Disable swap read-ahead
vm.page-cluster = 0
I figured that the readahead from the disk device
(/sys/block/zram0/queue/read_ahead_kb) is not meaningful---am I
correct?
These numbers are from a Chromebook with a few dozen Chrome tabs and a
couple of Android apps, and pretty heavy use of zram.
pgpgin 4688863
pgpgout 442052
pswpin 353675
pswpout 1072021
...
pgfault 5564247
pgmajfault 355758
pgmajfault_s 6297
pgmajfault_a 317645
pgmajfault_f 31816
pgmajfault_ax 8494
pgmajfault_fx 13201
where _s, _a, and _f are for shmem, anon, and file pages.
(ax and fx are for the subset of executable pages---I was curious about that)
So the numbers don't completely match:
anon faults = 318,000
swap ins = 354,000
Any idea of what might explain the difference?
> About auto resetting readahead with zram, I agree with you.
> But there are some reasons I postpone the work. No want to discuss
> it in this thread/moment. ;)
Yes, I wasn't even thinking of auto-resetting, just log a warning.
>> Incidentally, I understand anon and file faults, but what's a shmem fault?
>
> For me, it was out of my interest but if you want to count shmem fault,
> maybe, we need to introdue new stat(e.g., PSWPIN_SHM) in shmem_swapin
> but there are concrete reasons to justify in changelog. :)
Actually mine was a simpler question---I have no idea what a major
shmem fault is. And for this experiment it's a relatively small
number, but a similar order of magnitude to the (expensive) file
faults, so I don't want to completely ignore it.
Thanks again.
--
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-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
next prev parent reply other threads:[~2017-05-26 18:43 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-05-24 19:41 [PATCH] mm: add counters for different page fault types Luigi Semenzato
2017-05-25 0:19 ` Minchan Kim
2017-05-25 15:54 ` Luigi Semenzato
2017-05-26 4:06 ` Minchan Kim
2017-05-26 18:43 ` Luigi Semenzato [this message]
2017-05-29 8:15 ` Minchan Kim
2017-05-30 18:41 ` Luigi Semenzato
2017-05-31 1:31 ` Minchan Kim
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CAA25o9QG=Juynu-8wAYvdY1t7YNGVtE10fav2u3S-DikuU=aMQ@mail.gmail.com' \
--to=semenzato@google.com \
--cc=dianders@google.com \
--cc=dtor@google.com \
--cc=linux-mm@kvack.org \
--cc=minchan@kernel.org \
--cc=sonnyrao@google.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).