Projects/auto-ballooning: Difference between revisions

From KVM
No edit summary
 
(16 intermediate revisions by 2 users not shown)
Line 1: Line 1:
= Automatic Ballooning =
= Automatic Ballooning =


'''NOTE: This page describes an experimental development project from 2013 that was never completed. It is left here as a historical record. The feature described does not exist in any currently shipping version of QEMU'''
== Introduction ==
== Introduction ==


When a Linux host is running out of memory, the kernel will take action to ''reclaim'' memory. This action may be detrimental to KVM guests performace (eg. swapping) or even extreme to the point where the kernel may kill a VM or an important virt stack component.
The virtio balloon device allows KVM guests to reduce their memory size (thus relinquishing memory to the host) and to increase it back (thus taking memory from the host).


To help avoiding this scenario, a KVM guest could automatically give memory to the host when the host is facing memory pressure. By doing so the guest may also get into memory pressure so we need a way to allow the guest to automatically get memory back.
This feature is mainly intended to support over-committing memory on KVM hosts. That is, hosts that are running VMs whose total memory size is greater than what the host has physically available. For example, a 2G host running two VMs each with 2G would be over-committed.


== Design ==
The balloon device is important to support memory over-commitment because it allows for reducing a guest's memory size if needed. Suppose, in the previous example, that one of the guests is using 1G and its other 1G is free. We could use the balloon device to reduce this guest's size from 2G to 1G, this would free 1G in the host allowing the other VM to use it. Of course, if the reduced guest wants to run an application that consumes more than the 1G it currently has, it has to grow again.


KVM guests have a driver called the balloon driver. This driver allows guests to shrink and grow their memory. The balloon driver supports two operations:
That's the problem with the current balloon device, it's entirely manual. Someone (or some tool) is supposed to be watching the pressure in the host and guest and then operate the balloon accordingly. This doesn't work well in practice (if doable at all).


* Inflate: memory is taken from the guest and given to the host (guest shrinks)
The balloon has to be automatic in order to be really useful. It could like this: when the host is under pressure, it asks guests to relinquish some megas if they can. When/if a guest gets into memory pressure, it gets some megas back from the host. That's what the automatic ballooning series is about.
* Deflate: memory is returned from the host to the guest (guest grows)


Today, both operations are manual. The automatic ballooning project is about making them completely automatic, based on host and guest needs.
=== KVM Forum 2013 presentation slides ===


=== Automatic Inflate ===
[https://www.linux-kvm.org/images/5/58/Kvm-forum-2013-automatic-ballooning.pdf They can be found here], but note that much has changed since this talk.


Automatic inflate is performed by QEMU (ie. the KVM host). QEMU registers for [http://lwn.net/Articles/544652/ memory pressure] events so that it's notified when the host is under memory pressure.
== Patches and Git trees ==


Current patches have pre-defined values to be used by QEMU when it receives a memory pressure notification from the host kernel. Those values are:
=== Latest RFC version posted upstream ===


* 1MB on LOW pressure
'''Guest kernel:''' [http://marc.info/?l=kvm&m=138988948715638&w=2 RFC PATCH 0/4 virtio_balloon: add pressure notification via a new virtqueue]
* 2MB on MEDIUM pressure
* 4MB on CRITICAL pressure


For example, suppose the host is facing MEDIUM pressure and notifies QEMU. When QEMU receives the event, it asks the guest to inflate its balloon by 2MB. The guest in turn will shrink itself by 2MB and give that memory to the host.
'''QEMU:''' [http://marc.info/?l=kvm&m=138988966315690&w=2 RFC PATCH balloon: add automatic ballooning support]


=== Automatic Deflate ===
=== Git trees ===


Automatic deflate is performed by the guest or, more specifically, by the balloon driver.
'''Guest kernel:'''


The virtio-balloon driver registers a callback with the shrinker API. That callback is called when the guest kernel is facing memory pressure, and the number of pages to be returned to the kernel is passed to the callback. The balloon driver shrink callback deflates the guest's balloon.
[http://repo.or.cz/w/linux-2.6/luiz-linux-2.6.git git://repo.or.cz/linux-2.6/luiz-linux-2.6.git] virtio-balloon/pressure-notification/rfc/v1


For example, suppose the guest shrunk itself because of pressure in the host. But after some time, the guest is running more applications which causes memory pressure in the guest due its reduced size. The guest kernel then starts reclaiming memory and calls all shrink callbacks. That's when the balloon driver's shrink callback runs, and deflates the balloon by the number of pages specified. This causes the guest to grow again.
'''QEMU:'''


== GIT trees ==
[http://repo.or.cz/w/qemu/qmp-unstable.git git://repo.or.cz/qemu/qmp-unstable.git] balloon-automatic/handle-all-events/rfc/v1


Latest RFC version posted upstream:
== TODO ==


* QEMU: git://repo.or.cz/qemu/qmp-unstable.git balloon/auto-ballooning/rfc.v2 (or grab the patch [http://repo.or.cz/w/qemu/qmp-unstable.git/commit/59fec94e6396d6c32cdce47777440c3a988be63d here])
* More performance tests
* Guest kernel: git://repo.or.cz/linux-2.6/luiz-linux-2.6.git virtio-balloon/auto-deflate/rfc (or grab the first two patches from the [http://repo.or.cz/w/linux-2.6/luiz-linux-2.6.git/shortlog/refs/heads/virtio-balloon/auto-deflate/rfc web interface])
* Make it dynamic to enable/disable automatic ballooning
* Do kernel work required in order to have non-cgroups code using in-kernel memory pressure notification


== Testing ==
== Testing ==


You have to do three things to play with automatic-ballooning:
Setup:


# Install kernel 3.10 or higher in the host. Make sure to enable CONFIG_CGROUPS and CONFIG_MEMCG
# Install kernel 3.10 or higher in your host. Make sure the kernel options CONFIG_CGROUPS and CONFIG_MEMCG are enabled
# Clone QEMU from ''git://repo.or.cz/qemu/qmp-unstable.git balloon/auto-ballooning/rfc.v2'' (or grab the patch [http://repo.or.cz/w/qemu/qmp-unstable.git/commit/59fec94e6396d6c32cdce47777440c3a988be63d here])
# Build QEMU from [[#Git trees| the Git trees section]]
# Install the following kernel on your guest ''git://repo.or.cz/linux-2.6/luiz-linux-2.6.git virtio-balloon/auto-deflate/rfc'' (or grab the first two patches from the [http://repo.or.cz/w/linux-2.6/luiz-linux-2.6.git/shortlog/refs/heads/virtio-balloon/auto-deflate/rfc web interface])
# Build and install the guest kernel from [[#Git trees| the Git trees section]]


After setting up the above, do the following to experiment with automatic deflate (which is easy to reproduce):
Then start qemu with:


# Pass ''-balloon virtio,auto-balloon=true'' when starting QEMU
<nowiki>  # qemu [...] -device virtio-balloon,automatic=true</nowiki>
# Wait for the guest to boot, then generate some memory pressure in the guest (say a kernel build with lots of jobs)
# Switch to QEMU's monitor and shrink guest memory (say from 1G to 200MB)
# Watch the guest increase its memory by running "info balloon" on QEMU's monitor


To see automatic inflate and automatic deflate in action, you can run several VMs in parallel doing some heavy memory workload. Make sure to over commit host memory. Say your host has 4GB, run 6 VMs with 1GB each.
To see automatic ballooning in action, do the following:


Another idea is to run one or more VMs in a memory constrained cgroup, although this will require some hacking on the current patches as they register for the root memory cgroup.
# Generate pressure in your host (like running several VMs in parallel); or connect to QMP and issue the ''balloon-inject-host-pressure'' command
# You can monitor automatic ballooning activity by issuing ''info balloon'' in QEMU's monitor (also, I have debug code enabled that will dump a lot of info to stdout when "info balloon" is issued)
# After the guest has shrinked a bit, you can generate pressure in the guest to see it increasing its size again

Latest revision as of 06:51, 23 August 2018

Automatic Ballooning

NOTE: This page describes an experimental development project from 2013 that was never completed. It is left here as a historical record. The feature described does not exist in any currently shipping version of QEMU

Introduction

The virtio balloon device allows KVM guests to reduce their memory size (thus relinquishing memory to the host) and to increase it back (thus taking memory from the host).

This feature is mainly intended to support over-committing memory on KVM hosts. That is, hosts that are running VMs whose total memory size is greater than what the host has physically available. For example, a 2G host running two VMs each with 2G would be over-committed.

The balloon device is important to support memory over-commitment because it allows for reducing a guest's memory size if needed. Suppose, in the previous example, that one of the guests is using 1G and its other 1G is free. We could use the balloon device to reduce this guest's size from 2G to 1G, this would free 1G in the host allowing the other VM to use it. Of course, if the reduced guest wants to run an application that consumes more than the 1G it currently has, it has to grow again.

That's the problem with the current balloon device, it's entirely manual. Someone (or some tool) is supposed to be watching the pressure in the host and guest and then operate the balloon accordingly. This doesn't work well in practice (if doable at all).

The balloon has to be automatic in order to be really useful. It could like this: when the host is under pressure, it asks guests to relinquish some megas if they can. When/if a guest gets into memory pressure, it gets some megas back from the host. That's what the automatic ballooning series is about.

KVM Forum 2013 presentation slides

They can be found here, but note that much has changed since this talk.

Patches and Git trees

Latest RFC version posted upstream

Guest kernel: RFC PATCH 0/4 virtio_balloon: add pressure notification via a new virtqueue

QEMU: RFC PATCH balloon: add automatic ballooning support

Git trees

Guest kernel:

git://repo.or.cz/linux-2.6/luiz-linux-2.6.git virtio-balloon/pressure-notification/rfc/v1

QEMU:

git://repo.or.cz/qemu/qmp-unstable.git balloon-automatic/handle-all-events/rfc/v1

TODO

  • More performance tests
  • Make it dynamic to enable/disable automatic ballooning
  • Do kernel work required in order to have non-cgroups code using in-kernel memory pressure notification

Testing

Setup:

  1. Install kernel 3.10 or higher in your host. Make sure the kernel options CONFIG_CGROUPS and CONFIG_MEMCG are enabled
  2. Build QEMU from the Git trees section
  3. Build and install the guest kernel from the Git trees section

Then start qemu with:

# qemu [...] -device virtio-balloon,automatic=true

To see automatic ballooning in action, do the following:

  1. Generate pressure in your host (like running several VMs in parallel); or connect to QMP and issue the balloon-inject-host-pressure command
  2. You can monitor automatic ballooning activity by issuing info balloon in QEMU's monitor (also, I have debug code enabled that will dump a lot of info to stdout when "info balloon" is issued)
  3. After the guest has shrinked a bit, you can generate pressure in the guest to see it increasing its size again