Discussion:
[PATCH 0/8][for -mm] mem_notify v6
(too old to reply)
KOSAKI Motohiro
2008-02-09 15:19:58 UTC
Permalink
Hi

The /dev/mem_notify is low memory notification device.
it can avoid swappness and oom by cooperationg with the user process.

the Linux Today article is very nice description. (great works by Jake =
Edge)
http://www.linuxworld.com/news/2008/020508-kernel.html

<quoted>
When memory gets tight, it is quite possible that applications have mem=
ory
allocated=97often caches for better performance=97that they could free.
After all, it is generally better to lose some performance than to face=
the
consequences of being chosen by the OOM killer.
But, currently, there is no way for a process to know that the kernel i=
s
feeling memory pressure.
The patch provides a way for interested programs to monitor the /dev/me=
m_notify
file to be notified if memory starts to run low.
</quoted>


You need not be annoyed by OOM any longer :)
please any comments!

patch list
[1/8] introduce poll_wait_exclusive() new API
[2/8] introduce wake_up_locked_nr() new API
[3/8] introduce /dev/mem_notify new device (the core of this
patch series)
[4/8] memory_pressure_notify() caller
[5/8] add new mem_notify field to /proc/zoneinfo
[6/8] (optional) fixed incorrect shrink_zone
[7/8] ignore very small zone for prevent incorrect low mem notif=
y.
[8/8] support fasync feature


related discussion:
--------------------------------------------------------------
LKML OOM notifications requirement discussion
http://www.gossamer-threads.com/lists/linux/kernel/832802?nohighlig=
ht=3D1#832802
OOM notifications patch [Marcelo Tosatti]
http://marc.info/?l=3Dlinux-kernel&m=3D119273914027743&w=3D2
mem notifications v3 [Marcelo Tosatti]
http://marc.info/?l=3Dlinux-mm&m=3D119852828327044&w=3D2
Thrashing notification patch [Daniel Spang]
http://marc.info/?l=3Dlinux-mm&m=3D119427416315676&w=3D2
mem notification v4
http://marc.info/?l=3Dlinux-mm&m=3D120035840523718&w=3D2
mem notification v5
http://marc.info/?l=3Dlinux-mm&m=3D120114835421602&w=3D2

Changelog
-------------------------------------------------
v5 -> v6 (by KOSAKI Motohiro)
o rebase to 2.6.24-mm1
o fixed thundering herd guard formula.

v4 -> v5 (by KOSAKI Motohiro)
o rebase to 2.6.24-rc8-mm1
o change display order of /proc/zoneinfo
o ignore very small zone
o support fcntl(F_SETFL, FASYNC)
o fixed some trivial bugs.

v3 -> v4 (by KOSAKI Motohiro)
o rebase to 2.6.24-rc6-mm1
o avoid wake up all.
o add judgement point to __free_one_page().
o add zone awareness.

v2 -> v3 (by Marcelo Tosatti)
o changes the notification point to happen whenever
the VM moves an anonymous page to the inactive list.
o implement notification rate limit.

v1(oom notify) -> v2 (by Marcelo Tosatti)
o name change
o notify timing change from just swap thrashing to
just before thrashing.
o also works with swapless device.
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel=
" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Jon Masters
2008-02-09 16:02:55 UTC
Permalink
Yo,

Interesting patch series (I am being yuppie and reading this thread =20
from my iPhone on a treadmill at the gym - so further comments later). =
=20
I think that this is broadly along the lines that I was thinking, but =20
this should be an RFC only patch series for now.

Some initial questions:

Where is the netlink interface? Polling an FD is so last century :)

What testing have you done?

Still, it is good to start with some code - eventually we might just =20
have a full reservation API created. Rik and I and others have bounced =
=20
ideas around for a while and I hope we can pitch in. I will play with =20
these patches later.

Jon.



On Feb 9, 2008, at 10:19, "KOSAKI Motohiro" <***@jp.fujitsu=
=2Ecom=20
Post by KOSAKI Motohiro
Hi
The /dev/mem_notify is low memory notification device.
it can avoid swappness and oom by cooperationg with the user process.
the Linux Today article is very nice description. (great works by =20
Jake Edge)
http://www.linuxworld.com/news/2008/020508-kernel.html
<quoted>
When memory gets tight, it is quite possible that applications have =20
memory
allocated=E2=80=94often caches for better performance=E2=80=94that th=
ey could fre=20
Post by KOSAKI Motohiro
e.
After all, it is generally better to lose some performance than to =20
face the
consequences of being chosen by the OOM killer.
But, currently, there is no way for a process to know that the =20
kernel is
feeling memory pressure.
The patch provides a way for interested programs to monitor the /dev/=
=20
Post by KOSAKI Motohiro
mem_notify
file to be notified if memory starts to run low.
</quoted>
You need not be annoyed by OOM any longer :)
please any comments!
patch list
[1/8] introduce poll_wait_exclusive() new API
[2/8] introduce wake_up_locked_nr() new API
[3/8] introduce /dev/mem_notify new device (the core of this
patch series)
[4/8] memory_pressure_notify() caller
[5/8] add new mem_notify field to /proc/zoneinfo
[6/8] (optional) fixed incorrect shrink_zone
[7/8] ignore very small zone for prevent incorrect low mem =20
notify.
[8/8] support fasync feature
--------------------------------------------------------------
LKML OOM notifications requirement discussion
http://www.gossamer-threads.com/lists/linux/kernel/832802?nohighli=
ght=3D1#832802
Post by KOSAKI Motohiro
OOM notifications patch [Marcelo Tosatti]
http://marc.info/?l=3Dlinux-kernel&m=3D119273914027743&w=3D2
mem notifications v3 [Marcelo Tosatti]
http://marc.info/?l=3Dlinux-mm&m=3D119852828327044&w=3D2
Thrashing notification patch [Daniel Spang]
http://marc.info/?l=3Dlinux-mm&m=3D119427416315676&w=3D2
mem notification v4
http://marc.info/?l=3Dlinux-mm&m=3D120035840523718&w=3D2
mem notification v5
http://marc.info/?l=3Dlinux-mm&m=3D120114835421602&w=3D2
Changelog
-------------------------------------------------
v5 -> v6 (by KOSAKI Motohiro)
o rebase to 2.6.24-mm1
o fixed thundering herd guard formula.
v4 -> v5 (by KOSAKI Motohiro)
o rebase to 2.6.24-rc8-mm1
o change display order of /proc/zoneinfo
o ignore very small zone
o support fcntl(F_SETFL, FASYNC)
o fixed some trivial bugs.
v3 -> v4 (by KOSAKI Motohiro)
o rebase to 2.6.24-rc6-mm1
o avoid wake up all.
o add judgement point to __free_one_page().
o add zone awareness.
v2 -> v3 (by Marcelo Tosatti)
o changes the notification point to happen whenever
the VM moves an anonymous page to the inactive list.
o implement notification rate limit.
v1(oom notify) -> v2 (by Marcelo Tosatti)
o name change
o notify timing change from just swap thrashing to
just before thrashing.
o also works with swapless device.
KOSAKI Motohiro
2008-02-09 16:33:49 UTC
Permalink
Hi
Post by Jon Masters
Interesting patch series (I am being yuppie and reading this thread
from my iPhone on a treadmill at the gym - so further comments later).
I think that this is broadly along the lines that I was thinking, but
this should be an RFC only patch series for now.
sorry, I fixed at next post.
Thank you.
welcome to any discussion.
Post by Jon Masters
Where is the netlink interface? Polling an FD is so last century :)
to be honest, I don't know anyone use netlink and why hope receive
low memory notify by netlink.

poll() is old way, but it works good enough.

and, netlink have a bit weak point.
end up, netlink philosophy is read/write model.

I afraid to many low-mem message queued in netlink buffer
at under heavy pressure.
it cause degrade memory pressure.
Post by Jon Masters
Still, it is good to start with some code - eventually we might just
have a full reservation API created. Rik and I and others have bounced
ideas around for a while and I hope we can pitch in. I will play with
these patches later.
Great.
Welcome to any idea and any discussion.
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Rik van Riel
2008-02-09 16:43:29 UTC
Permalink
On Sun, 10 Feb 2008 01:33:49 +0900
Post by KOSAKI Motohiro
Post by Jon Masters
Where is the netlink interface? Polling an FD is so last century :)
to be honest, I don't know anyone use netlink and why hope receive
low memory notify by netlink.
poll() is old way, but it works good enough.
More importantly, all gtk+ programs, as well as most databases and other
system daemons have a poll() loop as their main loop.

A file descriptor fits that main loop perfectly.
--
All rights reversed.
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
KOSAKI Motohiro
2008-02-09 16:49:41 UTC
Permalink
Hi Rik
Post by Rik van Riel
More importantly, all gtk+ programs, as well as most databases and other
system daemons have a poll() loop as their main loop.
not only gtk+, may be all modern GUI program :)
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Jonathan Corbet
2008-02-11 15:36:02 UTC
Permalink
the Linux Today article is very nice description. (great works by Jake Edge)
http://www.linuxworld.com/news/2008/020508-kernel.html
Just for future reference...the above-mentioned article is from LWN,
syndicated onto LinuxWorld. It has, so far as I know, never been near
Linux Today.

Glad you liked it, though :)

Thanks,

jon
KOSAKI Motohiro
2008-02-11 15:46:07 UTC
Permalink
Post by Jonathan Corbet
the Linux Today article is very nice description. (great works by Jake Edge)
http://www.linuxworld.com/news/2008/020508-kernel.html
Just for future reference...the above-mentioned article is from LWN,
syndicated onto LinuxWorld. It has, so far as I know, never been near
Linux Today.
Glad you liked it, though :)
Oops, sorry.
I had serious misunderstand ;-)

sorry, again.
and thank you for your helpful message.
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Paul Jackson
2008-02-17 14:49:06 UTC
Permalink
I just noticed this patchset, kosaki-san. It looks quite interesting;
my apologies for not commenting earlier.

I see mention somewhere that mem_notify is of particular interest to
embedded systems.

I have what seems, intuitively, a similar problem at the opposite
end of the world, on big-honkin NUMA boxes (hundreds or thousands of
CPUs, terabytes of main memory.) The problem there is often best
resolved if we can kill the offending task, rather than shrink its
memory footprint. The situation is that several compute intensive
multi-threaded jobs are running, each in their own dedicated cpuset.

If one of these jobs tries to use more memory than is available in
its cpuset, then

(1) we quickly loose any hope of that job continuing at the excellent
performance needed of it, and

(2) we rapidly get increased risk of that job starting to swap and
unintentionally impact shared resources (kernel locks, disk
channels, disk heads).

So we like to identify such jobs as soon as they begin to swap,
and kill them very very quickly (before the direct reclaim code
in mm/vmscan.c can push more than a few pages to the swap device.)

For a much earlier, unsuccessful, attempt to accomplish this, see:

[Patch] cpusets policy kill no swap
http://lkml.org/lkml/2005/3/19/148

Now, it may well be that we are too far apart to share any part of
a solution; one seldom uses the same technology to build a Tour de
France bicycle as one uses to build a Lockheed C-5A Galaxy heavy
cargo transport.

One clear difference is the policy of what action we desire to take
when under memory pressure: do we invite user space to free memory so
as to avoid the wrath of the oom killer, or do we go to the opposite
extreme, seeking a nearly instantant killing, faster than the oom
killer can even begin its search for a victim.

Another clear difference is the use of cpusets, which are a major and
vital part of administering the big NUMA boxes, and I presume are not
even compiled into embedded kernels (correct?). This difference maybe
unbridgeable ... these big NUMA systems require per-cpuset mechanisms,
whereas embedded may require builds without cpusets.

However ... there might be some useful cross pollination of ideas.

I see in the latest posts to your mem_notify patchset v6, responding
to comments by Andrew and Andi on Feb 12 and 13, that you decided to
think more about the design of this, so perhaps this is a good time
for some random ideas from myself, even though I'm clearly coming from
a quite different problem space in some ways.

1) You have a little bit of code in the kernel to throttle the
thundering herd problem. Perhaps this could be moved to user space
... one user daemon that is always notified of such memory pressure
alarms, and in turn notifies interested applications. This might
avoid the need to add poll_wait_exclusive() to the kernel. And it
moves any fussy details of how to tame the thundering herd out of
the kernel.

2) Another possible mechanism for communicating events from
the kernel to user space is inotify. For example, I added
the line:

fsnotify_modify(dentry); # dentry is current tasks cpuset

at an interesting spot in vmscan.c, and using inotify-tools
<inotify-tools.sourceforge.net/> could easily watch all cpusets
for these events from one user space daemon.

At this point, I have no idea whether this odd use of inotify
is better or worse than what your patchset has. However using
inotify did require less new kernel code, and with such user space
mechanisms as inotify-tools already well developed, it made the
problem I had, of watching an entire hierarcy of special files
(beneath /dev/cpuset) very easy to implement. At least inotify
also presents events on a file descriptor that can be consumed
using a poll() loop.

3) Perhaps, instead of sending simple events, one could update
a meter of the rate of recent such events, such as the per-cpuset
'memory_pressure' mechanism does. This might lead to addressing
Andrew Morton's comment:

If this feature is useful then I'd expect that some
applications would want notification at different times, or at
different levels of VM distress. So this semi-randomly-chosen
notification point just won't be strong enough in real-world
use.

4) A place that I found well suited for my purposes (watching for
swapping from direct reclaim) was just before the lines in the
pageout() routine in mm/vmscan.c:

if (clear_page_dirty_for_io(page)) {
...
res = mapping->a_ops->writepage(page, &wbc);

It seemed that testing "PageAnon(page)" here allowed me to easily
distinguish between dirty pages going back to the file system, and
pages going to swap (this detail is from work on a 2.6.16 kernel;
things might have changed.)

One possible advantage of the above hook in the direct reclaim
code path in vmscan.c is that pressure in one cpuset did not cause
any false alarms in other cpusets. However even this hook does
not take into account the constraints of mm/mempolicy (the NUMA
memory policy that Andi mentioned) nor of cgroup memory controllers.

5) I'd be keen to find an agreeable way that you could have the
system-wide, no cpuset, mechanism you need, while at the same
time, I have a cpuset interface that is similar and depends on the
same set of hooks. This might involve a single set of hooks in
the key places in the memory and swapping code, that (1) updated
the system wide state you need, and (2) if cpusets were present,
updated similar state for the tasks current cpuset. The user
visible API would present both the system-wide connector you need
(the special file or whatever) and if cpusets are present, similar
per-cpuset connectors.

Anyhow ... just some thoughts. Perhaps one of them will be useful.
--
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <***@sgi.com> 1.940.382.4214
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
KOSAKI Motohiro
2008-02-19 07:36:23 UTC
Permalink
Hi Paul,

Thank you for wonderful interestings comment.
your comment is really nice.

I was HPC guy with large NUMA box at past.
I promise i don't ignroe hpc user.
but unfortunately I didn't have experience of use CPUSET
because at that point, it was under development yet.

I hope discuss you that CPUSET usage case and mem_notify requirement.
to be honest, I thought hpc user doesn't use mem_notify, sorry.
Post by Paul Jackson
I have what seems, intuitively, a similar problem at the opposite
end of the world, on big-honkin NUMA boxes (hundreds or thousands of
CPUs, terabytes of main memory.) The problem there is often best
resolved if we can kill the offending task, rather than shrink its
memory footprint. The situation is that several compute intensive
multi-threaded jobs are running, each in their own dedicated cpuset.
agreed.
Post by Paul Jackson
So we like to identify such jobs as soon as they begin to swap,
and kill them very very quickly (before the direct reclaim code
in mm/vmscan.c can push more than a few pages to the swap device.)
you think kill the process just after swap, right?
but unfortunately, almost user hope receive notification before swap ;-)
because avoid swap.

I think we need discuss this point more.
Post by Paul Jackson
[Patch] cpusets policy kill no swap
http://lkml.org/lkml/2005/3/19/148
Now, it may well be that we are too far apart to share any part of
a solution; one seldom uses the same technology to build a Tour de
France bicycle as one uses to build a Lockheed C-5A Galaxy heavy
cargo transport.
One clear difference is the policy of what action we desire to take
when under memory pressure: do we invite user space to free memory so
as to avoid the wrath of the oom killer, or do we go to the opposite
extreme, seeking a nearly instantant killing, faster than the oom
killer can even begin its search for a victim.
Hmm, sorry
I understand your patch yet, because I don't know CPUSET so much.

I learn CPUSET more, about this week and I'll reply again about next week ;-)
Post by Paul Jackson
Another clear difference is the use of cpusets, which are a major and
vital part of administering the big NUMA boxes, and I presume are not
even compiled into embedded kernels (correct?). This difference maybe
unbridgeable ... these big NUMA systems require per-cpuset mechanisms,
whereas embedded may require builds without cpusets.
Yes, some embedded distribution(i.e. monta vista) distribute as source.
but embedded people strongly dislike bloat code size.
I think they never turn on CPUSET.

I hope mem_notify works fine without CPUSET.
Post by Paul Jackson
1) You have a little bit of code in the kernel to throttle the
thundering herd problem. Perhaps this could be moved to user space
... one user daemon that is always notified of such memory pressure
alarms, and in turn notifies interested applications. This might
avoid the need to add poll_wait_exclusive() to the kernel. And it
moves any fussy details of how to tame the thundering herd out of
the kernel.
I think you talk about user space oom manager.
it and many user process are obviously different.

I doubt memory manager daemon model doesn't works on desktop and
typical server.
thus, current implementaion optimize to no manager environment.

of course, it doesn't mean i refuse add to code for oom manager.
it is very interesting idea.

i hope discussion it more.
Post by Paul Jackson
2) Another possible mechanism for communicating events from
the kernel to user space is inotify. For example, I added
fsnotify_modify(dentry); # dentry is current tasks cpuset
Excellent!
that is really good idea.

thaks.
Post by Paul Jackson
3) Perhaps, instead of sending simple events, one could update
a meter of the rate of recent such events, such as the per-cpuset
'memory_pressure' mechanism does. This might lead to addressing
If this feature is useful then I'd expect that some
applications would want notification at different times, or at
different levels of VM distress. So this semi-randomly-chosen
notification point just won't be strong enough in real-world
use.
Hmmm, I don't think so.
I think timing of memmory_pressure_notify(1) is already best.

the page move active list to inactive list indicate swap I/O happen
a bit after.

but memmory_pressure_notify(0) is a bit messy.
I'll try to improve more simplify.
Post by Paul Jackson
4) A place that I found well suited for my purposes (watching for
swapping from direct reclaim) was just before the lines in the
if (clear_page_dirty_for_io(page)) {
...
res = mapping->a_ops->writepage(page, &wbc);
It seemed that testing "PageAnon(page)" here allowed me to easily
distinguish between dirty pages going back to the file system, and
pages going to swap (this detail is from work on a 2.6.16 kernel;
things might have changed.)
One possible advantage of the above hook in the direct reclaim
code path in vmscan.c is that pressure in one cpuset did not cause
any false alarms in other cpusets. However even this hook does
not take into account the constraints of mm/mempolicy (the NUMA
memory policy that Andi mentioned) nor of cgroup memory controllers.
Disagreed.
that is too late.

after writepage notifify mean can't avoid swap I/O.
Post by Paul Jackson
5) I'd be keen to find an agreeable way that you could have the
system-wide, no cpuset, mechanism you need, while at the same
time, I have a cpuset interface that is similar and depends on the
same set of hooks. This might involve a single set of hooks in
the key places in the memory and swapping code, that (1) updated
the system wide state you need, and (2) if cpusets were present,
updated similar state for the tasks current cpuset. The user
visible API would present both the system-wide connector you need
(the special file or whatever) and if cpusets are present, similar
per-cpuset connectors.
that makes sense.
I will learn cpuset and think integrate mem_notify and cpuset.


and,

Please don't think I reject your idea.
your proposal is large different of past our discussion and
i don't know cpuset.

I think we can't drop all current design and accept your idea all, may be.
but we may be able to accept partial until hpc guys content enough.

I will learn to CPUSET more in a few days.
after it, we can discussion more.

please wait for a while.

Thanks!



-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Paul Jackson
2008-02-19 15:00:08 UTC
Permalink
Post by KOSAKI Motohiro
Thank you for wonderful interestings comment.
You're most welcome. The pleasure is all mine.
Post by KOSAKI Motohiro
you think kill the process just after swap, right?
but unfortunately, almost user hope receive notification before swap ;-)
because avoid swap.
There is not much my customers HPC jobs can do with notification before
swap. Their jobs either have the main memory they need to perform the
requested calculations with the desired performance, or their job is
useless and should be killed. Unlike the applications you describe,
my customers jobs have no way, once running, to adapt to less memory.
They can only adapt to less memory by being restarted with a different
set of resource requests to the job scheduler (the application that
manages job requests, assigns them CPU, memory and other resources,
and monitors, starts, stops and pauses jobs.)

The primary difficulty my HPC customers have is killing such jobs fast
enough, before a bad job (one that attempts to use more memory than it
signed up for) can harm the performance of other users and the rest of
the system.

I don't mind if a pages are slowly or occassionally written to swap;
but as soon as the task wants to reclaim big chunks of memory by
writing thousands of pages at once to swap, it must die, and die
before it can queue more than a handful of those pages to the swapper.
Post by KOSAKI Motohiro
but embedded people strongly dislike bloat code size.
I think they never turn on CPUSET.
I hope mem_notify works fine without CPUSET.
Yes - understood and agreed - as I guessed, cpusets are not configured
in embedded systems.
Post by KOSAKI Motohiro
Please don't think I reject your idea.
your proposal is large different of past our discussion
Yes - I agree that my ideas were quite different. Please don't
hesitate to reject every one of them, like a Samurai slicing through
air with his favorite sword <grin>.
Post by KOSAKI Motohiro
Disagreed. that [my direct reclaim hook at mapping->a_ops->writepage()]
is too late.
For your work, yes that hook is too late. Agreed.

Depending on what we're trying to do:
1) warn applications of swap coming soon (your case),
2) show how close we are to swapping,
3) show how much swap has happened already,
4) kill instantly if try to swap (my hpc case),
5) measure file i/o caused by memory pressure, or
6) perhaps other goals,
we will need to hook different places in the kernel.

It may well be that your hooks for embedded are simply in different
places than my hooks for HPC. If so, that's fine.

I look forward to your further thoughts.
--
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <***@sgi.com> 1.940.382.4214
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Rik van Riel
2008-02-19 19:02:22 UTC
Permalink
On Tue, 19 Feb 2008 09:00:08 -0600
Post by Paul Jackson
1) warn applications of swap coming soon (your case),
2) show how close we are to swapping,
3) show how much swap has happened already,
4) kill instantly if try to swap (my hpc case),
5) measure file i/o caused by memory pressure, or
6) perhaps other goals,
we will need to hook different places in the kernel.
It may well be that your hooks for embedded are simply in different
places than my hooks for HPC. If so, that's fine.
Don't forget the "hooks for desktop" :)

Basically in all situations, the kernel needs to warn at the same point
in time: when the system is about to run out of RAM for anonymous pages.

In the desktop case, that leads to swapping (and programs can free memory).

In the embedded case, it leads to OOM (and a management program can kill or
restart something else, or a program can restart itself).

In the HPC case, it leads to swapping (and a management program can kill or
restart something else).

I do not see the kernel side being different between these situations, only
userspace reacts differently in the different scenarios.

Am I overlooking something?
--
All Rights Reversed
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Paul Jackson
2008-02-19 20:18:20 UTC
Permalink
Post by Rik van Riel
Basically in all situations, the kernel needs to warn at the same point
in time: when the system is about to run out of RAM for anonymous pages.
...
In the HPC case, it leads to swapping (and a management program can kill or
restart something else).
Thanks for stopping by ...

Perhaps with the cgroup based memory controller in progress, or with
other work I'm overlooking, this is or will no longer be a problem,
but on 2.6.16 kernels (the latest ones I have in major production HPC
use) this is not sufficient.

As of at least that point, we don't (didn't ?) have sufficiently
accurate numbers of when we were "about to run out". We can only
detect when "we just did run out", as evidenced by entering the direct
reclaim code, or as by slightly later events such as starting to push
Anon pages to the swap device from direct reclaim.

Actually, even the point that we enter direct reclaim, near the bottom
of __alloc_pages(), isn't adequate either, as we could be there because
some thread in that cpuset is trying to write out a results file that
is larger than that cpusets memory. In that case, we really don't want
to kill the job ... it just needs to be (and routinely is) throttled
back to disk speeds as it completes the write out of dirty file system
pages.

So the first clear spot that we -know- serious swapping is commencing
is where the direct reclaim code calls a writepage op with an Anon
page. At that point, having a management program intervene is entirely
too late. Even having the task at that instant, inline, tag itself
with a SIGKILL, as it queues that first Anon page to a swap device, is
too late. The direct reclaim code can loop, pushing hundreds or
thousand of pages, on big memory systems, to the swapper, in the
current reclaim loop, before it pops the stack far enough back to even
notice that it has a SIGKILL pending on it. The suppression of pushing
pages to the swapper has to happen right there, inline in some
mm/vmscan.c code, as part of the direct reclaim loops.

(Hopefully I said something stupid in that last paragraph, and you will
be able to correct it ... it sure would be useful ;).

A year or two ago, I added the 'memory_pressure' per-cpuset meter to
Linux, in an effort to realize just what you suggest, Rik. My colleagues
at SGI (mostly) and myself (a little) have proven to ourselves that this
doesn't work, for our HPC needs, for two reasons:

1) once swapping begins, issuing a SIGKILL, no matter how instantly,
is too late, as explained above, and

2) that memory_pressure combines and confuses memory pressure due to
dirty file system buffers filling memory, with memory pressure due
to anonymous swappable pages filling memory, also as explained above.

I do have a patch in my playarea that adds two more of the
memory_pressure meters, one for swapouts, and one for flushing dirty
file system buffers, both hooking into the spot in the vmscan reclaim
code where the writepage op is called. This patch ~might~ address the
desktop need here. It nicely generates two clean, sharp indicators
that we're getting throttled by direct reclaim of dirty file buffers,
and that we're starting to reclaim anon pages to the swappers. Of course
for embedded use, I'd have to adapt it to a non-cpuset based mechanism
(not difficult), as embedded definitely doesn't do cpusets.
Post by Rik van Riel
Don't forget the "hooks for desktop" :)
I'll agree (perhaps out of ignorance) that the desktop (and normal sized
server) cases are like the embedded case ... in that they need to
distribute some event to user space tasks that want to know that memory
is short, so that that user space code can do what it will (reclaim
some user space memory or restart or kill or throttle something?)

However, I'm still stuck carrying a patch out of the community kernel,
to get my HPC customers the "instant kill on direct reclaim swap" they
need, as this still seems to be the special case. Which is rather
unfortunate, from the business perspective of my good employer, as it
the -only- out of mainline patch, so far as I know, that we have been
having to carry, continuously, for several years now. But for that
single long standing issue (and now and then various more short term
issues), a vanilla distribution kernel, using the vanilla distribution
config and build, runs production on our one or two thousand CPU,
several terabyte big honkin NUMA boxes.

Part of my motivation for engaging Kosaki-san in this discussion was
to reinvigorate this discussion, as it's getting to be time I took
another shot at getting something in the community kernel that addresses
this. The more overlap the HPC fix here is with the other 99.978% of
the worlds Linux systems that are desktop, laptop, (ordinary sized)
server or embedded, the better my chances (he said hopefully.)

(Now if I could just get us to consider systems in proportion to
how much power & cooling they need, rather than in proportion to
unit sales ... ;)
--
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <***@sgi.com> 1.940.382.4214
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Paul Jackson
2008-02-19 20:43:29 UTC
Permalink
Post by Paul Jackson
Of course
for embedded use, I'd have to adapt it to a non-cpuset based mechanism
(not difficult), as embedded definitely doesn't do cpusets.
I'm forgetting an important detail here. Kosaki-san has clearly stated
that this hook, at vmscan's writepage, is too late for his embedded needs,
and that they need the feedback a bit earlier, when the page moves from
the active list to the inactive list.

However, except for the placement of such hooks in three or four
places, rather than just one, it may well be (if cpusets could be
factored out) that one mechanism would meet all needs ... except for
that pesky HPC need for throttling to more or less zero the swapping
from select cpusets.
--
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <***@sgi.com> 1.940.382.4214
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Pavel Machek
2008-02-19 22:28:28 UTC
Permalink
Post by Paul Jackson
Post by KOSAKI Motohiro
Thank you for wonderful interestings comment.
You're most welcome. The pleasure is all mine.
Post by KOSAKI Motohiro
you think kill the process just after swap, right?
but unfortunately, almost user hope receive notification before swap ;-)
because avoid swap.
There is not much my customers HPC jobs can do with notification before
swap. Their jobs either have the main memory they need to perform the
requested calculations with the desired performance, or their job is
useless and should be killed. Unlike the applications you describe,
my customers jobs have no way, once running, to adapt to less
memory.
Sounds like a job for memory limits (ulimit?), not for OOM
notification, right?
Pavel
--
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Paul Jackson
2008-02-20 01:54:49 UTC
Permalink
Post by Pavel Machek
Post by Paul Jackson
There is not much my customers HPC jobs can do with notification before
swap. Their jobs either have the main memory they need to perform the
requested calculations with the desired performance, or their job is
useless and should be killed. Unlike the applications you describe,
my customers jobs have no way, once running, to adapt to less memory.
Sounds like a job for memory limits (ulimit?), not for OOM
notification, right?
Er eh -- which one?

The only one I see that might help keep a multi-threaded job
using various kinds of memory on multiple nodes confined could
be the resident set size (RLIMIT_RSS; ulimit -m). So far as
I can tell, that one is a pure no-op in Linux.

Here's the bash list of all available ulimit (setrlimit) options:

-a All current limits are reported
-c The maximum size of core files created
-d The maximum size of a process's data segment
-e The maximum scheduling priority ("nice")
-f The maximum size of files written by the shell and its children
-i The maximum number of pending signals
-l The maximum size that may be locked into memory
-m The maximum resident set size
-n The maximum number of open file descriptors (most systems do not allow this value to be set)
-p The pipe size in 512-byte blocks (this may not be set)
-q The maximum number of bytes in POSIX message queues
-r The maximum real-time scheduling priority
-s The maximum stack size
-t The maximum amount of cpu time in seconds
-u The maximum number of processes available to a single user
-v The maximum amount of virtual memory available to the shell
-x The maximum number of file locks

Did I miss seeing one that would be useful?

Actually, given the chronic problem we've had over the years accounting
for how much memory in total (including text, data, stack, mapped
files, locked pages, kernel memory structures that an application is
using many of, ... I'd be suprised if any such ulimit existed that
actually worked for this purpose (confining an HPC jobs to using almost
exactly all the memory available to it, but no more.)
--
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <***@sgi.com> 1.940.382.4214
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Rik van Riel
2008-02-20 02:07:39 UTC
Permalink
On Tue, 19 Feb 2008 23:28:28 +0100
Post by Pavel Machek
Sounds like a job for memory limits (ulimit?), not for OOM
notification, right?
I suspect one problem could be that an HPC job scheduling program
does not know exactly how much memory each job can take, so it can
sometimes end up making a mistake and overcommitting the memory on
one HPC node.

In that case the user is better off having that job killed and
restarted elsewhere, than having all of the jobs on that node
crawl to a halt due to swapping.

Paul, is this guess correct? :)
--
All rights reversed.
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
KOSAKI Motohiro
2008-02-20 02:48:41 UTC
Permalink
Hi Rik
Post by Rik van Riel
Post by Pavel Machek
Sounds like a job for memory limits (ulimit?), not for OOM
notification, right?
I suspect one problem could be that an HPC job scheduling program
does not know exactly how much memory each job can take, so it can
sometimes end up making a mistake and overcommitting the memory on
one HPC node.
In that case the user is better off having that job killed and
restarted elsewhere, than having all of the jobs on that node
crawl to a halt due to swapping.
Paul, is this guess correct? :)
Yes.
Fujitsu HPC middleware watching sum of memory consumption of the job
and, if over-consumption happened, kill process and remove job schedule.

I think that is common hpc requirement.
but we watching to user defined memory limit, not swap.

Thanks.


-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Paul Jackson
2008-02-20 04:57:33 UTC
Permalink
Post by KOSAKI Motohiro
Yes.
Fujitsu HPC middleware watching sum of memory consumption of the job
and, if over-consumption happened, kill process and remove job schedule.
Did those jobs share nodes -- sometimes two or more jobs using the same
nodes? I am sure SGI has such users too, though such job mixes make
the runtimes of specific jobs less obvious, so customers are more
tolerant of variations and some inefficiencies, as they get hidden in
the mix.

In other words, Rik, both yes and no ;). Both sorts of HPC loads
exist, sharing nodes and a dedicated set of nodes for each job.
--
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <***@sgi.com> 1.940.382.4214
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
KOSAKI Motohiro
2008-02-20 05:21:37 UTC
Permalink
Post by Paul Jackson
Did those jobs share nodes -- sometimes two or more jobs using the same
nodes? I am sure SGI has such users too, though such job mixes make
the runtimes of specific jobs less obvious, so customers are more
tolerant of variations and some inefficiencies, as they get hidden in
the mix.
Hm
our dedicated ndoe user set memory limit to machine physical memory
size (minus a bit).

I think don't have so much share/dedicate and watch user-defined/swap.
am i misundestand?


-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Paul Jackson
2008-02-20 04:36:25 UTC
Permalink
Post by Rik van Riel
In that case the user is better off having that job killed and
restarted elsewhere, than having all of the jobs on that node
crawl to a halt due to swapping.
Paul, is this guess correct? :)
Not for the loads I focus on. Each job gets exclusive use of its own
dedicated set of nodes, for the duration of the job. With that comes a
quite specific upper limit on how much memory, in total, including node
local kernel data, that job is allowed to use.

One problem with swapping is that nodes aren't entirely isolated.
They share buses, i/o channels, disk arms, kernel data cache lines and
kernel locks with other nodes, running other jobs. A job thrashing
its swap is a drag on the rest of the system.

Another problem with swapping is that it's a waste of resources. Once
a pure compute bound job goes into swapping when it shouldn't, that job
has near zero hope of continuing with the intended performance, as it
has just slowed from main memory speeds to disk speeds, which are
thousands of times slower. Best to get it out of there, immediately.
--
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <***@sgi.com> 1.940.382.4214
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Tom May
2008-04-01 23:35:06 UTC
Permalink
On Sat, Feb 9, 2008 at 8:19 AM, KOSAKI Motohiro
Post by KOSAKI Motohiro
Hi
The /dev/mem_notify is low memory notification device.
it can avoid swappness and oom by cooperationg with the user process=
=2E
Post by KOSAKI Motohiro
the Linux Today article is very nice description. (great works by Ja=
ke Edge)
Post by KOSAKI Motohiro
http://www.linuxworld.com/news/2008/020508-kernel.html
<quoted>
When memory gets tight, it is quite possible that applications have =
memory
Post by KOSAKI Motohiro
allocated=97often caches for better performance=97that they could fr=
ee.
Post by KOSAKI Motohiro
After all, it is generally better to lose some performance than to f=
ace the
Post by KOSAKI Motohiro
consequences of being chosen by the OOM killer.
But, currently, there is no way for a process to know that the kerne=
l is
Post by KOSAKI Motohiro
feeling memory pressure.
The patch provides a way for interested programs to monitor the /dev=
/mem_notify
Post by KOSAKI Motohiro
file to be notified if memory starts to run low.
</quoted>
You need not be annoyed by OOM any longer :)
please any comments!
Thanks for this patch set! I ported it to 2.6.23.9 and tried it, on a
system with no swap since I'm evaluating this for an embedded system.
In practice, the criterion it uses for notifications wasn't sufficient =
to avoid
memory problems, including OOM, in a cyclic allocate/notify/free
sequence which is probably typical.

I tried it with a real-world program that, among other things, mmaps
anonymous pages and touches them at a reasonable speed until it gets
notified via /dev/mem_notify, releases most of them with
madvise(MADV_DONTNEED), then loops to start the cycle again.

What tends to happen is that I do indeed get notifications via
/dev/mem_notify when the kernel would like to be swapping, at which
point I free memory. But the notifications come at a time when the
kernel needs memory, and it gets the memory by discarding some Cached
or Mapped memory (I can see these decreasing in /proc/meminfo with
each notification). With each mmap/notify/madvise cycle the Cached
and Mapped memory gets smaller, until eventually while I'm touching
pages the kernel can't find enough memory and will either invoke the
OOM killer or return ENOMEM from syscalls. This is precisely the
situation I'm trying to avoid by using /dev/mem_notify.

The criterion of "notify when the kernel would like to swap" feels
correct, but in addition I seem to need something like "notify when
cached+mapped+free memory is getting low".

I'll need to be looking into doing this, so any comments or ideas are
welcome.

Thanks,
=2Etom
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel=
" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
KOSAKI Motohiro
2008-04-02 07:31:27 UTC
Permalink
Hi Tom,

Thank you very useful comment.
that is very interesting.
Post by Tom May
I tried it with a real-world program that, among other things, mmaps
anonymous pages and touches them at a reasonable speed until it gets
notified via /dev/mem_notify, releases most of them with
madvise(MADV_DONTNEED), then loops to start the cycle again.
What tends to happen is that I do indeed get notifications via
/dev/mem_notify when the kernel would like to be swapping, at which
point I free memory. But the notifications come at a time when the
kernel needs memory, and it gets the memory by discarding some Cached
or Mapped memory (I can see these decreasing in /proc/meminfo with
each notification). With each mmap/notify/madvise cycle the Cached
and Mapped memory gets smaller, until eventually while I'm touching
pages the kernel can't find enough memory and will either invoke the
OOM killer or return ENOMEM from syscalls. This is precisely the
situation I'm trying to avoid by using /dev/mem_notify.
Could you send your test program?
I can't reproduce that now, sorry.
Post by Tom May
The criterion of "notify when the kernel would like to swap" feels
correct, but in addition I seem to need something like "notify when
cached+mapped+free memory is getting low".
Hmmm,
I think this idea is only useful when userland process call
madvise(MADV_DONTNEED) periodically.

but I hope improve my patch and solve your problem.
if you don' mind, please help my testing ;)
Tom May
2008-04-02 17:45:05 UTC
Permalink
On Wed, Apr 2, 2008 at 12:31 AM, KOSAKI Motohiro
Post by KOSAKI Motohiro
Hi Tom,
Thank you very useful comment.
that is very interesting.
Post by Tom May
I tried it with a real-world program that, among other things, mmaps
anonymous pages and touches them at a reasonable speed until it gets
notified via /dev/mem_notify, releases most of them with
madvise(MADV_DONTNEED), then loops to start the cycle again.
What tends to happen is that I do indeed get notifications via
/dev/mem_notify when the kernel would like to be swapping, at which
point I free memory. But the notifications come at a time when the
kernel needs memory, and it gets the memory by discarding some Cached
or Mapped memory (I can see these decreasing in /proc/meminfo with
each notification). With each mmap/notify/madvise cycle the Cached
and Mapped memory gets smaller, until eventually while I'm touching
pages the kernel can't find enough memory and will either invoke the
OOM killer or return ENOMEM from syscalls. This is precisely the
situation I'm trying to avoid by using /dev/mem_notify.
Could you send your test program?
Unfortunately, no, it's a Java Virtual Machine (which is a perfect
user of /dev/mem_notify since it can garbage collect on notification,
among other times).

But it should be possible to make a small program with the same
behavior; I'll do that.
Post by KOSAKI Motohiro
I can't reproduce that now, sorry.
Post by Tom May
The criterion of "notify when the kernel would like to swap" feels
correct, but in addition I seem to need something like "notify when
cached+mapped+free memory is getting low".
Hmmm,
I think this idea is only useful when userland process call
madvise(MADV_DONTNEED) periodically.
Do you have a recommendation for freeing memory? I could maybe use
munmap/mmap, but that's not atomic and may be "worse" (more overhead,
etc.) than madvise(MADV_DONTNEED).
Post by KOSAKI Motohiro
but I hope improve my patch and solve your problem.
if you don' mind, please help my testing ;)
It's my pleasure to help in any way I can.

.tom
Tom May
2008-04-15 00:16:50 UTC
Permalink
On Wed, Apr 2, 2008 at 12:31 AM, KOSAKI Motohiro
Post by KOSAKI Motohiro
Hi Tom,
Thank you very useful comment.
that is very interesting.
Post by Tom May
I tried it with a real-world program that, among other things, mmaps
anonymous pages and touches them at a reasonable speed until it gets
notified via /dev/mem_notify, releases most of them with
madvise(MADV_DONTNEED), then loops to start the cycle again.
What tends to happen is that I do indeed get notifications via
/dev/mem_notify when the kernel would like to be swapping, at which
point I free memory. But the notifications come at a time when the
kernel needs memory, and it gets the memory by discarding some Cached
or Mapped memory (I can see these decreasing in /proc/meminfo with
each notification). With each mmap/notify/madvise cycle the Cached
and Mapped memory gets smaller, until eventually while I'm touching
pages the kernel can't find enough memory and will either invoke the
OOM killer or return ENOMEM from syscalls. This is precisely the
situation I'm trying to avoid by using /dev/mem_notify.
Could you send your test program?
I can't reproduce that now, sorry.
Post by Tom May
The criterion of "notify when the kernel would like to swap" feels
correct, but in addition I seem to need something like "notify when
cached+mapped+free memory is getting low".
Hmmm,
I think this idea is only useful when userland process call
madvise(MADV_DONTNEED) periodically.
but I hope improve my patch and solve your problem.
if you don' mind, please help my testing ;)
Here's a test program that allocates memory and frees on notification.
It takes an argument which is the number of pages to use; use a
number considerably higher than the amount of memory in the system.
I'm running this on a system without swap. Each time it gets a
notification, it frees memory and writes out the /proc/meminfo
contents. What I see is that Cached gradually decreases, then Mapped
decreases, and eventually the kernel invokes the oom killer. It may
be necessary to tune some of the constants that control the allocation
and free rates and latency; these values work for my system.

#include <stdio.h>
#include <stdlib.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <poll.h>
#include <sched.h>
#include <time.h>

#define PAGESIZE 4096

/* How many pages we've mmap'd. */
static int pages;

/* Pointer to mmap'd memory used as a circular buffer. One thread
touches pages, another thread releases them on notification. */
static char *p;

/* How many pages to touch each 5ms. This makes at most 2000
pages/sec. */
#define TOUCH_CHUNK 10

/* How many pages to free when we're notified. With a 100ms FREE_DELAY,
we can free ~9110 pages/sec, or perhaps only 5*911 = 4555 pages/sec if we're
notified only 5 times/sec. */
#define FREE_CHUNK 911

/* Delay in milliseconds before freeing pages, to simulate latency while finding
pages to free. */
#define FREE_DELAY 100

static void touch(void);
static int release(void *arg);
static void release_pages(void);
static void show_meminfo(void);

/* Stack for the release thread. */
static char stack[8192];

int
main (int argc, char **argv)
{
pages = atoi(argv[1]);
p = mmap(NULL, pages * PAGESIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE, 0, 0);
if (p == MAP_FAILED) {
perror("mmap");
exit(1);
}

if (clone(release, stack + sizeof(stack) - 4,
CLONE_VM | CLONE_FS | CLONE_FILES | CLONE_SIGHAND | CLONE_THREAD,
NULL) == -1) {
perror("clone failed");
exit(1);
}

touch();
}

static void
touch (void)
{
int page = 0;

while (1) {
int i;
struct timespec t;
for (i = 0; i < TOUCH_CHUNK; i++) {
p[page * PAGESIZE] = 1;
if (++page >= pages) {
page = 0;
}
}

t.tv_sec = 0;
t.tv_nsec = 5000000;
if (nanosleep(&t, NULL) == -1) {
perror("nanosleep");
}
}
}

static int
release (void *arg)
{
int fd = open("/dev/mem_notify", O_RDONLY);
if (fd == -1) {
perror("open(/dev/mem_notify)");
exit(1);
}

while (1) {
struct pollfd pfd;
int nfds;

pfd.fd = fd;
pfd.events = POLLIN;

nfds = poll(&pfd, 1, -1);
if (nfds == -1) {
perror("poll");
exit(1);
}
if (nfds == 1) {
struct timespec t;
t.tv_sec = 0;
t.tv_nsec = FREE_DELAY * 1000000;
if (nanosleep(&t, NULL) == -1) {
perror("nanosleep");
}
release_pages();
printf("time: %d\n", time(NULL));
show_meminfo();
}
}
}

static void
release_pages (void)
{
/* Index of the next page to free. */
static int page = 0;
int i;

/* Release FREE_CHUNK pages. */

for (i = 0; i < FREE_CHUNK; i++) {
int r = madvise(p + page*PAGESIZE, PAGESIZE, MADV_DONTNEED);
if (r == -1) {
perror("madvise");
exit(1);
}
if (++page >= pages) {
page = 0;
}
}
}

static void
show_meminfo (void)
{
char buffer[2000];
int fd;
ssize_t n;

fd = open("/proc/meminfo", O_RDONLY);
if (fd == -1) {
perror("open(/proc/meminfo)");
exit(1);
}

n = read(fd, buffer, sizeof(buffer));
if (n == -1) {
perror("read(/proc/meminfo)");
exit(1);
}

n = write(1, buffer, n);
if (n == -1) {
perror("write(stdout)");
exit(1);
}

if (close(fd) == -1) {
perror("close(/proc/meminfo)");
exit(1);
}
}

.tom
KOSAKI Motohiro
2008-04-16 02:30:55 UTC
Permalink
Post by Tom May
Here's a test program that allocates memory and frees on notification.
It takes an argument which is the number of pages to use; use a
number considerably higher than the amount of memory in the system.
I'm running this on a system without swap. Each time it gets a
notification, it frees memory and writes out the /proc/meminfo
contents. What I see is that Cached gradually decreases, then Mapped
decreases, and eventually the kernel invokes the oom killer. It may
be necessary to tune some of the constants that control the allocation
and free rates and latency; these values work for my system.
really thanks!
I'll test soon :)
KOSAKI Motohiro
2008-04-17 09:30:11 UTC
Permalink
Hi Tom
Post by Tom May
Here's a test program that allocates memory and frees on notification.
It takes an argument which is the number of pages to use; use a
number considerably higher than the amount of memory in the system.
I'm running this on a system without swap. Each time it gets a
notification, it frees memory and writes out the /proc/meminfo
contents. What I see is that Cached gradually decreases, then Mapped
decreases, and eventually the kernel invokes the oom killer. It may
be necessary to tune some of the constants that control the allocation
and free rates and latency; these values work for my system.
may be...

I think you misunderstand madvise(MADV_DONTNEED).
madvise(DONTNEED) indicate drop process page table.
it mean become easily swap.

when run on system without swap, madvise(DONTNEED) almost doesn't work
as your expected.

I am sorry for being not able to help you. ;)
Tom May
2008-04-17 19:23:20 UTC
Permalink
On Thu, Apr 17, 2008 at 2:30 AM, KOSAKI Motohiro
Post by KOSAKI Motohiro
Hi Tom
Post by Tom May
Here's a test program that allocates memory and frees on notification.
It takes an argument which is the number of pages to use; use a
number considerably higher than the amount of memory in the system.
I'm running this on a system without swap. Each time it gets a
notification, it frees memory and writes out the /proc/meminfo
contents. What I see is that Cached gradually decreases, then Mapped
decreases, and eventually the kernel invokes the oom killer. It may
be necessary to tune some of the constants that control the allocation
and free rates and latency; these values work for my system.
may be...
I think you misunderstand madvise(MADV_DONTNEED).
madvise(DONTNEED) indicate drop process page table.
it mean become easily swap.
when run on system without swap, madvise(DONTNEED) almost doesn't work
as your expected.
madvise can be replaced with munmap and the same behavior occurs.

--- test.c.orig 2008-04-17 11:41:47.000000000 -0700
+++ test.c 2008-04-17 11:44:04.000000000 -0700
@@ -127,7 +127,7 @@
/* Release FREE_CHUNK pages. */

for (i = 0; i < FREE_CHUNK; i++) {
- int r = madvise(p + page*PAGESIZE, PAGESIZE, MADV_DONTNEED);
+ int r = munmap(p + page*PAGESIZE, PAGESIZE);
if (r == -1) {
perror("madvise");
exit(1);

Here's what I'm seeing on my system. This is with munmap, but I see
the same thing with madvise. First, /proc/meminfo on my system before
running the test:

# cat /proc/meminfo
MemTotal: 127612 kB
MemFree: 71348 kB
Buffers: 1404 kB
Cached: 52324 kB
SwapCached: 0 kB
Active: 2336 kB
Inactive: 51656 kB
SwapTotal: 0 kB
SwapFree: 0 kB
Dirty: 80 kB
Writeback: 0 kB
AnonPages: 276 kB
Mapped: 376 kB
Slab: 1680 kB
SReclaimable: 824 kB
SUnreclaim: 856 kB
PageTables: 52 kB
NFS_Unstable: 0 kB
Bounce: 0 kB
CommitLimit: 63804 kB
Committed_AS: 908 kB
VmallocTotal: 909280 kB
VmallocUsed: 304 kB
VmallocChunk: 908976 kB

Here is the start and end of the output from the test program. At
each /dev/mem_notify notification Cached decreases, then eventually
Mapped decreases as well, which means the amount of time the program
has to free memory gets smaller and smaller. Finally the oom killer
is invoked because the program can't react quickly enough to free
memory, even though it can free at a faster rate than it can use
memory. My test is slow to free because it calls nanosleep, but this
is just a simulation of my actual program that has to perform garbage
collection before it can free memory.

# ./test_unmap 250000
time: 1208458019
MemTotal: 127612 kB
MemFree: 5524 kB
Buffers: 872 kB
Cached: 18388 kB
SwapCached: 0 kB
Active: 101468 kB
Inactive: 18220 kB
SwapTotal: 0 kB
SwapFree: 0 kB
Dirty: 0 kB
Writeback: 0 kB
AnonPages: 100436 kB
Mapped: 504 kB
Slab: 1608 kB
SReclaimable: 816 kB
SUnreclaim: 792 kB
PageTables: 152 kB
NFS_Unstable: 0 kB
Bounce: 0 kB
CommitLimit: 63804 kB
Committed_AS: 944 kB
VmallocTotal: 909280 kB
VmallocUsed: 304 kB
VmallocChunk: 908976 kB

time: 1208458020
MemTotal: 127612 kB
MemFree: 5732 kB
Buffers: 820 kB
Cached: 17928 kB
SwapCached: 0 kB
Active: 101708 kB
Inactive: 17752 kB
SwapTotal: 0 kB
SwapFree: 0 kB
Dirty: 0 kB
Writeback: 0 kB
AnonPages: 100712 kB
Mapped: 504 kB
Slab: 1608 kB
SReclaimable: 816 kB
SUnreclaim: 792 kB
PageTables: 156 kB
NFS_Unstable: 0 kB
Bounce: 0 kB
CommitLimit: 63804 kB
Committed_AS: 944 kB
VmallocTotal: 909280 kB
VmallocUsed: 304 kB
VmallocChunk: 908976 kB

time: 1208458021
MemTotal: 127612 kB
MemFree: 5660 kB
Buffers: 820 kB
Cached: 17416 kB
SwapCached: 0 kB
Active: 102228 kB
Inactive: 17316 kB
SwapTotal: 0 kB
SwapFree: 0 kB
Dirty: 0 kB
Writeback: 0 kB
AnonPages: 101308 kB
Mapped: 504 kB
Slab: 1608 kB
SReclaimable: 816 kB
SUnreclaim: 792 kB
PageTables: 156 kB
NFS_Unstable: 0 kB
Bounce: 0 kB
CommitLimit: 63804 kB
Committed_AS: 944 kB
VmallocTotal: 909280 kB
VmallocUsed: 304 kB
VmallocChunk: 908976 kB

--- snip --- now Mapped is decreasing: ---

time: 1208458049
MemTotal: 127612 kB
MemFree: 5568 kB
Buffers: 40 kB
Cached: 868 kB
SwapCached: 0 kB
Active: 119036 kB
Inactive: 720 kB
SwapTotal: 0 kB
SwapFree: 0 kB
Dirty: 0 kB
Writeback: 0 kB
AnonPages: 118848 kB
Mapped: 488 kB
Slab: 1456 kB
SReclaimable: 724 kB
SUnreclaim: 732 kB
PageTables: 172 kB
NFS_Unstable: 0 kB
Bounce: 0 kB
CommitLimit: 63804 kB
Committed_AS: 944 kB
VmallocTotal: 909280 kB
VmallocUsed: 304 kB
VmallocChunk: 908976 kB

time: 1208458050
MemTotal: 127612 kB
MemFree: 5608 kB
Buffers: 40 kB
Cached: 356 kB
SwapCached: 0 kB
Active: 119392 kB
Inactive: 328 kB
SwapTotal: 0 kB
SwapFree: 0 kB
Dirty: 0 kB
Writeback: 0 kB
AnonPages: 119324 kB
Mapped: 268 kB
Slab: 1456 kB
SReclaimable: 724 kB
SUnreclaim: 732 kB
PageTables: 172 kB
NFS_Unstable: 0 kB
Bounce: 0 kB
CommitLimit: 63804 kB
Committed_AS: 944 kB
VmallocTotal: 909280 kB
VmallocUsed: 304 kB
VmallocChunk: 908976 kB

time: 1208458051
MemTotal: 127612 kB
MemFree: 5428 kB
Buffers: 40 kB
Cached: 116 kB
SwapCached: 0 kB
Active: 119832 kB
Inactive: 84 kB
SwapTotal: 0 kB
SwapFree: 0 kB
Dirty: 0 kB
Writeback: 0 kB
AnonPages: 119760 kB
Mapped: 60 kB
Slab: 1440 kB
SReclaimable: 720 kB
SUnreclaim: 720 kB
PageTables: 172 kB
NFS_Unstable: 0 kB
Bounce: 0 kB
CommitLimit: 63804 kB
Committed_AS: 944 kB
VmallocTotal: 909280 kB
VmallocUsed: 304 kB
VmallocChunk: 908976 kB
test_unmap invoked oom-killer: gfp_mask=0xa80d2, order=0, oomkilladj=0
[<c012f1db>] out_of_memory+0x16f/0x1a0
[<c01308dd>] __alloc_pages+0x2c1/0x300
[<c013757a>] handle_mm_fault+0x262/0x3e4
[<c010906b>] do_page_fault+0x407/0x638
[<c011fba0>] hrtimer_wakeup+0x0/0x18
[<c0108c64>] do_page_fault+0x0/0x638
[<c024d822>] error_code+0x6a/0x70

If it's possible to get a notification when MemFree + Cached + Mapped
(I'm not sure whether this is the right formula) falls below some
threshold, so that the program has time to find memory to discard
before the system runs out, that would prevent the oom -- as long as
the application(s) can ensure that there is not too much memory
allocated while it is looking for memory to free. But at least the
threshold would give it a reasonable amount of time to handle the
notification.

Thanks,
.tom
KOSAKI Motohiro
2008-04-18 10:07:49 UTC
Permalink
Post by Tom May
madvise can be replaced with munmap and the same behavior occurs.
--- test.c.orig 2008-04-17 11:41:47.000000000 -0700
+++ test.c 2008-04-17 11:44:04.000000000 -0700
@@ -127,7 +127,7 @@
/* Release FREE_CHUNK pages. */
for (i = 0; i < FREE_CHUNK; i++) {
- int r = madvise(p + page*PAGESIZE, PAGESIZE, MADV_DONTNEED);
+ int r = munmap(p + page*PAGESIZE, PAGESIZE);
if (r == -1) {
perror("madvise");
exit(1);
Here's what I'm seeing on my system. This is with munmap, but I see
the same thing with madvise. First, /proc/meminfo on my system before
Oh sorry my bad!
I investigated again and found 2 problem in your test program.

1. text segment isn't locked.

if strong memory pressure happned, kernel may drop program text region.
then your test program suddenly slow down.

please use mlockall(MCL_CURRENT) before large buffer allocation.

2. repeat open/close to /proc/meminfo.

in the fact, open(2) system call use a bit memory.
if call open(2) in strong memory pressure, doesn't return until
memory freed enough.
thus, it cause slow down your program sometimes.

attached changed test program :)
it works well on my test environment.
Post by Tom May
If it's possible to get a notification when MemFree + Cached + Mapped
(I'm not sure whether this is the right formula) falls below some
threshold, so that the program has time to find memory to discard
before the system runs out, that would prevent the oom -- as long as
the application(s) can ensure that there is not too much memory
allocated while it is looking for memory to free. But at least the
threshold would give it a reasonable amount of time to handle the
notification.
your proposal is interesting.
but I hope try to my attached test program at first.
Tom May
2008-04-21 20:32:06 UTC
Permalink
On Fri, Apr 18, 2008 at 3:07 AM, KOSAKI Motohiro
Post by KOSAKI Motohiro
I investigated again and found 2 problem in your test program.
1. text segment isn't locked.
if strong memory pressure happned, kernel may drop program text region.
then your test program suddenly slow down.
please use mlockall(MCL_CURRENT) before large buffer allocation.
Using mlock does enable the program to respond faster (and/or the
kernel doesn't have to find memory to fault the page in) and solves
the problem for this simple test program. I think we're thinking of
the solution in two different ways: you want the program to react more
quickly or be "nicer", and I want the kernel to give notification
early enough to allow time for things that can (and do) happen when
things aren't so nice. I realize that in extreme circumstances oom may
be unavoidable, but a threshold-based notification, in addition to the
current /dev/mem_notify mechanism, would help avoid extreme
circumstances. I'm going to look into doing this.
Post by KOSAKI Motohiro
2. repeat open/close to /proc/meminfo.
in the fact, open(2) system call use a bit memory.
if call open(2) in strong memory pressure, doesn't return until
memory freed enough.
thus, it cause slow down your program sometimes.
This should be fine; I intentionally do the open/read/write/close
after freeing memory.
Post by KOSAKI Motohiro
attached changed test program :)
it works well on my test environment.
I made your changes to my program (I'm using clone since I don't have
a pthreads library on my device, and I left PAGESIZE at 4K instead of
64K), and having memory locked does avoid oom in this case, but
unfortunately I don't think it's a general solution that will work
everywhere in my system. (Although I'm going to try it.)

Thanks,
.tom
Tom May
2008-05-01 02:07:58 UTC
Permalink
Post by KOSAKI Motohiro
Hi Tom
Here is the start and end of the output from the test program. A=
t
Post by KOSAKI Motohiro
each /dev/mem_notify notification Cached decreases, then eventual=
ly
Post by KOSAKI Motohiro
Mapped decreases as well, which means the amount of time the prog=
ram
Post by KOSAKI Motohiro
has to free memory gets smaller and smaller. Finally the oom kil=
ler
Post by KOSAKI Motohiro
is invoked because the program can't react quickly enough to free
memory, even though it can free at a faster rate than it can use
memory. My test is slow to free because it calls nanosleep, but =
this
Post by KOSAKI Motohiro
is just a simulation of my actual program that has to perform gar=
bage
Post by KOSAKI Motohiro
collection before it can free memory.
I have also seen this behaviour in my static tests with low mem
notification on swapless systems. It is a problem with small program=
s
Post by KOSAKI Motohiro
(typically static test programs) where the text segment is only a fe=
w
Post by KOSAKI Motohiro
pages. I have not seen this behaviour in larger programs which use a
larger working set. As long as the system working set is bigger than
the amount of memory that needs to be allocated, between every
notification reaction opportunity, it seems to be ok.
Hi Daniel,

You're saying the program's in-core text pages serve as a reserve that
the kernel can discard when it needs some memory, correct? And that
even if the kernel discards them, it will page them back in as a
matter of course as the program runs, to maintain the reserve? That
certainly makes sense.

In my case of a Java virtual machine, where I originally saw the
problem, most of the code is interpreted byte codes or jit-compiled
native code, all of which resides not in the text segment but in
anonymous pages that aren't backed by a file, and there is no swap
space. The actual text segment working set can be very small (memory
allocation, garbage collection, synchronization, other random native
code). And, as KOSAKI Motohiro pointed out, it may be wise to mlock
these areas. So the text working set doesn't make an adequate
reserve.

However, I can maintain a reserve of cached and/or mapped memory by
touching pages in the text segment (or any mapped file) as the final
step of low memory notification handling, if the cached page count is
getting low. For my purposes, this is nearly the same as having an
additional threshold-based notification, since it forces notifications
to occur while the kernel still has some memory to satisfy allocations
while userspace code works to free memory. And it's simple.

Unfortunately, this is more expensive than it could be since the pages
need to be read in from some device (mapping /dev/zero doesn't cause
pages to be allocated). What I'm looking for now is a cheap way to
populate the cache with pages that the kernel can throw away when it
needs to reclaim memory.

Thanks,
=2Etom
KOSAKI Motohiro
2008-05-01 15:06:05 UTC
Permalink
Hi Tom,
Post by Tom May
In my case of a Java virtual machine, where I originally saw the
problem, most of the code is interpreted byte codes or jit-compiled
native code, all of which resides not in the text segment but in
anonymous pages that aren't backed by a file, and there is no swap
space. The actual text segment working set can be very small (memory
allocation, garbage collection, synchronization, other random native
code). And, as KOSAKI Motohiro pointed out, it may be wise to mlock
these areas. So the text working set doesn't make an adequate
reserve.
your memnotify check routine is written by native or java?
if native, my suggestion is right.
but if java, it is wrong.

my point is "on swapless system, /dev/mem_notify checked routine should be mlocked".
Post by Tom May
However, I can maintain a reserve of cached and/or mapped memory by
touching pages in the text segment (or any mapped file) as the final
step of low memory notification handling, if the cached page count is
getting low. For my purposes, this is nearly the same as having an
additional threshold-based notification, since it forces notifications
to occur while the kernel still has some memory to satisfy allocations
while userspace code works to free memory. And it's simple.
Unfortunately, this is more expensive than it could be since the pages
need to be read in from some device (mapping /dev/zero doesn't cause
pages to be allocated). What I'm looking for now is a cheap way to
populate the cache with pages that the kernel can throw away when it
needs to reclaim memory.
I hope understand your requirement more.
Can I ask your system more?

I think all java text and data is mapped.
When cached+mapped+free memory is happend?
and at the time, What is used memory?

Please don't think I have objection your proposal.
merely, I don't understand your system yet.

if I make new code before understand your requirement exactly,
It makes many bug.


IMHO threshold based notification has a problems.
if low memory happend and application has no freeable memory,
mem notification don't stop and increase CPU usage dramatically, but it is perfectly useless.

I don't thin embedded java is not important, but I don't hope
desktop regression...
Tom May
2008-05-02 22:21:40 UTC
Permalink
On Thu, May 1, 2008 at 8:06 AM, KOSAKI Motohiro
Post by KOSAKI Motohiro
Hi Tom,
Post by Tom May
In my case of a Java virtual machine, where I originally saw the
problem, most of the code is interpreted byte codes or jit-compiled
native code, all of which resides not in the text segment but in
anonymous pages that aren't backed by a file, and there is no swap
space. The actual text segment working set can be very small (memory
allocation, garbage collection, synchronization, other random native
code). And, as KOSAKI Motohiro pointed out, it may be wise to mlock
these areas. So the text working set doesn't make an adequate
reserve.
your memnotify check routine is written by native or java?
Some of each.
Post by KOSAKI Motohiro
if native, my suggestion is right.
but if java, it is wrong.
my point is "on swapless system, /dev/mem_notify checked routine should be mlocked".
mlocking didn't fix things, it just made the oom happen at a different
time (see graphs below), both in the small test program where I used
mlockall, and in the jvm where during initialization I read
/proc/self/maps and mlocked each region of memory that was mapped to a
file. Note that without swap, all of the anonymous pages containing
the java code are effectively locked in memory, too, so everything
runs without page faults.
Post by KOSAKI Motohiro
Post by Tom May
However, I can maintain a reserve of cached and/or mapped memory by
touching pages in the text segment (or any mapped file) as the final
step of low memory notification handling, if the cached page count is
getting low. For my purposes, this is nearly the same as having an
additional threshold-based notification, since it forces notifications
to occur while the kernel still has some memory to satisfy allocations
while userspace code works to free memory. And it's simple.
Unfortunately, this is more expensive than it could be since the pages
need to be read in from some device (mapping /dev/zero doesn't cause
pages to be allocated). What I'm looking for now is a cheap way to
populate the cache with pages that the kernel can throw away when it
needs to reclaim memory.
I hope understand your requirement more.
Most simply, I need to get low memory notifications while there is
still enough memory to handle them before oom.
Post by KOSAKI Motohiro
Can I ask your system more?
x86, Linux 2.6.23.9 (with your patches trivially backported), 128MB,
no swap. Is there something else I can tell you?
Post by KOSAKI Motohiro
I think all java text and data is mapped.
It's not what /proc/meminfo calls "Mapped". It's in anonymous pages
with no backing store, i.e., mmap with MAP_ANONYMOUS.
Post by KOSAKI Motohiro
When cached+mapped+free memory is happend?
and at the time, What is used memory?
Here's a graph of MemFree, Cached, and Mapped over time (I believe
Mapped is mostly or entirely subset of Cached here, so it's not
actually important):

Loading Image...

The amount of MemFree fluctuates as java allocates and garbage
collects, but the Cached memory decreases (since the kernel has to use
it for __alloc_pages when memory is low) until at some point there is
no memory to satisfy __alloc_pages and there is an oom.

The same things happens if I use mlock, only it happens sooner because
the kernel can't discard any of the 15MB of mlock'd memory so it
actually runs out of memory faster:

Loading Image...

I'm not sure how to explain it differently than I have before. Maybe
someone else could explain it better. So, at the risk of merely
repeating myself: The jvm allocates memory. MemFree decreases. In
the kernel, __alloc_pages is called. It finds that memory is low,
memory_pressure_notify is called, and some cached pages are moved to
the inactive list. These pages may then be used to satisfy
__alloc_pages requests. The jvm gets the notification, collects
garbage, and returns memory to the kernel which appears as MemFree in
/proc/meminfo. The cycle continues: the jvm allocates memory until
memory_pressure_notify is called, more cached pages are moved to the
inactive list, etc. Eventually there are no more pages to move to the
inactive list, and __alloc_pages will invoke the oom killer.
Post by KOSAKI Motohiro
Please don't think I have objection your proposal.
merely, I don't understand your system yet.
if I make new code before understand your requirement exactly,
It makes many bug.
Of course.
Post by KOSAKI Motohiro
IMHO threshold based notification has a problems.
if low memory happend and application has no freeable memory,
mem notification don't stop and increase CPU usage dramatically, but it is perfectly useless.
My thought was to notify only when the threshold is crossed, i.e.,
edge-triggered not level-triggered. But I now think a threshold
mechanism may be too complicated, and artificially putting pages in
the cache works just as well. As a proof-of-concept, I do this, and
it works well, but is inefficient:

extern char _text;
for (int i = 0; i < bytes; i += 4096) {
*((volatile char *)&_text + i);
}
Post by KOSAKI Motohiro
I don't thin embedded java is not important, but I don't hope
desktop regression...
I think embedded Java is a perfect user of /dev/mem_notify :-) I was
happy to see your patches and the criteria you used for notification.
But I'm having a problem in practice :-(

.tom
KOSAKI Motohiro
2008-05-03 12:26:49 UTC
Permalink
Post by Tom May
Post by KOSAKI Motohiro
your memnotify check routine is written by native or java?
Some of each.
Wow!
you have 2 /dev/mem_notify checking routine?
java routine free java memory, native routine free native memory, right?
Post by Tom May
Post by KOSAKI Motohiro
my point is "on swapless system, /dev/mem_notify checked routine should be mlocked".
mlocking didn't fix things, it just made the oom happen at a different
time (see graphs below), both in the small test program where I used
mlockall, and in the jvm where during initialization I read
/proc/self/maps and mlocked each region of memory that was mapped to a
file. Note that without swap, all of the anonymous pages containing
the java code are effectively locked in memory, too, so everything
runs without page faults.
okey.
Post by Tom May
Post by KOSAKI Motohiro
I hope understand your requirement more.
Most simply, I need to get low memory notifications while there is
still enough memory to handle them before oom.
Ah, That's your implementation idea.
I hope know why don't works well my current implementation at first.
Post by Tom May
Post by KOSAKI Motohiro
Can I ask your system more?
x86, Linux 2.6.23.9 (with your patches trivially backported), 128MB,
no swap. Is there something else I can tell you?
Post by KOSAKI Motohiro
I think all java text and data is mapped.
It's not what /proc/meminfo calls "Mapped". It's in anonymous pages
with no backing store, i.e., mmap with MAP_ANONYMOUS.
okey.
Mapped of /proc/meminfo mean mapped pages with file backing store.

therefore, that isn't contain anonymous memory(e.g. java).
Post by Tom May
Post by KOSAKI Motohiro
When cached+mapped+free memory is happend?
and at the time, What is used memory?
Here's a graph of MemFree, Cached, and Mapped over time (I believe
Mapped is mostly or entirely subset of Cached here, so it's not
http://www.tommay.net/memory.png
I hope know your system memory usage detail.
your system have 128MB, but your graph vertical line represent 0M - 35M.
Who use remain 93MB(128-35)?
We should know who use memory intead decrease cached memory.

So, Can you below operation before mesurement?

# echo 100 > /proc/sys/vm/swappiness
# echo 3 >/proc/sys/vm/drop_caches

and, Can you mesure AnonPages of /proc/meminfo too?
(Can your memory shrinking routine reduce anonymous memory?)

if JVM use memory as anonymous memory and your memory shrinking routine can't
anonymous memory, that isn't mem_notify proble,
that is just poor JVM garbege collection problem.

Why I think that?
mapped page of your graph decrease linearly.
if notification doesn't happened, it doesn't decrease.

thus,
in your system, memory notification is happend rightly.
but your JVM doesn't have enough freeable memory.

if my assumption is right, increase number of memory notification
doesn't solve your problem.

Sould we find way of good interaction to JVM GC and mem_notify shrinker?
Sould mem_notify shrinker kick JVM GC for shrink anonymous memory?
Post by Tom May
My thought was to notify only when the threshold is crossed, i.e.,
edge-triggered not level-triggered.
Hm, interesting..
Post by Tom May
But I now think a threshold
mechanism may be too complicated, and artificially putting pages in
the cache works just as well. As a proof-of-concept, I do this, and
extern char _text;
for (int i = 0; i < bytes; i += 4096) {
*((volatile char *)&_text + i);
}
you intent populate to .text segment?
if so, you can mamp(MAP_POPULATE), IMHO.
Post by Tom May
I think embedded Java is a perfect user of /dev/mem_notify :-) I was
happy to see your patches and the criteria you used for notification.
But I'm having a problem in practice :-(
Yeah, absolutely.
I'll try to set up JVM to my test environment tomorrow.
Tom May
2008-05-06 05:22:03 UTC
Permalink
All the talk about how what is using memory, whether things are in
Java or native, etc., is missing the point. I'm not sure I'm making
my point, so I'll try again: regardless of memory size, mlock, etc.,
the interaction between client and /dev/mem_notify can be unstable,
and is unstable in my case:

- User-space code page faults in MAP_ANONYMOUS regions until there is
no free memory.
- The kernel gives a notification.
- There kernel frees some cache to satisfy the memory request.
- The user-space code gets the notification and frees anonymous pages.
Concurrently
with this, some thread(s) in the system may continue to page fault.
- The cycle repeats.
- This works well, perhaps hundreds or thousands of cycles, until all
or most of the cache has been freed and we get an oom handling a page
fault.

My requirement is to have a stable system, with memory allocated on
demand to whatever process(es) want it (jvm, web browser, ...) until a
low memory notification occurs, which causes them to free whatever
memory they no longer need, then continue, without arbitrary static
limits on Java heap size, web browser cache size, etc.

My workaround to make things stable is to put pages in the cache
(after releasing anonymous pages and increasing free memory) by
accessing pages in _text, but that seems silly and expensive.

.tom

On Sat, May 3, 2008 at 5:26 AM, KOSAKI Motohiro
Post by KOSAKI Motohiro
Post by Tom May
Post by KOSAKI Motohiro
your memnotify check routine is written by native or java?
Some of each.
Wow!
you have 2 /dev/mem_notify checking routine?
java routine free java memory, native routine free native memory, right?
Post by Tom May
Post by KOSAKI Motohiro
my point is "on swapless system, /dev/mem_notify checked routine should be mlocked".
mlocking didn't fix things, it just made the oom happen at a different
time (see graphs below), both in the small test program where I used
mlockall, and in the jvm where during initialization I read
/proc/self/maps and mlocked each region of memory that was mapped to a
file. Note that without swap, all of the anonymous pages containing
the java code are effectively locked in memory, too, so everything
runs without page faults.
okey.
Post by Tom May
Post by KOSAKI Motohiro
I hope understand your requirement more.
Most simply, I need to get low memory notifications while there is
still enough memory to handle them before oom.
Ah, That's your implementation idea.
I hope know why don't works well my current implementation at first.
Post by Tom May
Post by KOSAKI Motohiro
Can I ask your system more?
x86, Linux 2.6.23.9 (with your patches trivially backported), 128MB,
no swap. Is there something else I can tell you?
Post by KOSAKI Motohiro
I think all java text and data is mapped.
It's not what /proc/meminfo calls "Mapped". It's in anonymous pages
with no backing store, i.e., mmap with MAP_ANONYMOUS.
okey.
Mapped of /proc/meminfo mean mapped pages with file backing store.
therefore, that isn't contain anonymous memory(e.g. java).
Post by Tom May
Post by KOSAKI Motohiro
When cached+mapped+free memory is happend?
and at the time, What is used memory?
Here's a graph of MemFree, Cached, and Mapped over time (I believe
Mapped is mostly or entirely subset of Cached here, so it's not
http://www.tommay.net/memory.png
I hope know your system memory usage detail.
your system have 128MB, but your graph vertical line represent 0M - 35M.
Who use remain 93MB(128-35)?
We should know who use memory intead decrease cached memory.
So, Can you below operation before mesurement?
# echo 100 > /proc/sys/vm/swappiness
# echo 3 >/proc/sys/vm/drop_caches
and, Can you mesure AnonPages of /proc/meminfo too?
(Can your memory shrinking routine reduce anonymous memory?)
if JVM use memory as anonymous memory and your memory shrinking routine can't
anonymous memory, that isn't mem_notify proble,
that is just poor JVM garbege collection problem.
Why I think that?
mapped page of your graph decrease linearly.
if notification doesn't happened, it doesn't decrease.
thus,
in your system, memory notification is happend rightly.
but your JVM doesn't have enough freeable memory.
if my assumption is right, increase number of memory notification
doesn't solve your problem.
Sould we find way of good interaction to JVM GC and mem_notify shrinker?
Sould mem_notify shrinker kick JVM GC for shrink anonymous memory?
Post by Tom May
My thought was to notify only when the threshold is crossed, i.e.,
edge-triggered not level-triggered.
Hm, interesting..
Post by Tom May
But I now think a threshold
mechanism may be too complicated, and artificially putting pages in
the cache works just as well. As a proof-of-concept, I do this, and
extern char _text;
for (int i = 0; i < bytes; i += 4096) {
*((volatile char *)&_text + i);
}
you intent populate to .text segment?
if so, you can mamp(MAP_POPULATE), IMHO.
Post by Tom May
I think embedded Java is a perfect user of /dev/mem_notify :-) I was
happy to see your patches and the criteria you used for notification.
But I'm having a problem in practice :-(
Yeah, absolutely.
I'll try to set up JVM to my test environment tomorrow.
Continue reading on narkive:
Loading...