From: Andrea Righi <righi.andrea@gmail.com>
To: balbir@linux.vnet.ibm.com
Cc: menage@google.com, kamezawa.hiroyu@jp.fujitsu.com,
       kosaki.motohiro@jp.fujitsu.com, xemul@openvz.org,
       linux-kernel@vger.kernel.org, containers@lists.osdl.org
Subject: [RFC PATCH 0/5] memcg: VM overcommit accounting and handling
Date: Tue, 10 Jun 2008 01:32:58 +0200
Message-Id: <1213054383-18137-1-git-send-email-righi.andrea@gmail.com>
In-Reply-To: <>
References: <>
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2000
Lines: 54


Provide distinct cgroup VM overcommit accounting and handling using the memory
resource controller.

Patchset against latest Linus git tree.

This patchset allows to set different per-cgroup overcommit rules and,
according to them, it's possible to return a memory allocation failure (ENOMEM)
to the applications, instead of always triggering the OOM killer via
mem_cgroup_out_of_memory() when cgroup memory limits are exceeded.

Default overcommit settings are taken from vm.overcommit_memory and
vm.overcommit_ratio sysctl values. Child cgroups initially inherits the VM
overcommit parent's settings.

Cgroup overcommit settings can be overridden using memory.overcommit_memory and
memory.overcommit_ratio files under the cgroup filesystem.

For example:

1. Initialize a cgroup with 50MB memory limit:
# mount -t cgroup none /cgroups -o memory
# mkdir /cgroups/0
# /bin/echo $$ >  /cgroups/0/tasks
# /bin/echo 50M > /cgroups/0/memory.limit_in_bytes

2. Use the "never overcommit" policy with 50% ratio:
# /bin/echo 2 > /cgroups/0/memory.overcommit_memory
# /bin/echo 50 > /cgroups/0/memory.overcommit_ratio

Assuming we have no swap space, cgroup 0 can allocate up to 25MB of virtual
memory. If that limit is exceeded all the further allocation attempts made by
userspace applications will receive a -ENOMEM.

4. Show committed VM statistics:
# cat /cgroups/0/memory.overcommit_as
CommitLimit:     25600 kB
Committed_AS:     9844 kB

5. Use "always overcommmit":
# /bin/echo 1 > /cgroups/0/memory.overcommit_memory

This is very similar to the default memory controller configuration: overcommit
is allowed, but when there's no more available memory oom-killer is invoked.

TODO:
- shared memory is not taken in account (i.e. files in tmpfs)

-Andrea
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/