Comments on: Using CGroups with libvirt and LXC/KVM guests in Fedora 12 http://berrange.com/posts/2009/12/03/using-cgroups-with-libvirt-and-lxckvm-guests-in-fedora-12/ Writing about photography, open source software, virtualization & more Sun, 13 May 2012 09:58:15 +0000 hourly 1 http://wordpress.org/?v=3.3.2 By: Vineet http://berrange.com/posts/2009/12/03/using-cgroups-with-libvirt-and-lxckvm-guests-in-fedora-12/comment-page-1/#comment-25272 Vineet Tue, 01 May 2012 20:49:58 +0000 http://wordpress.berrange.com/?p=12#comment-25272 Hi Daniel, I've use cgroups on CentOS and I've noticed the cgroup hierarchy for libvirt that you mentioned. However, when I install all the packages for kubuntu, I noticed that the same structure does not get auto-mounted. I checked libvirtd.conf/qemu.conf and cgconfig.conf, all look same. Is there something i'm missing to get the same hierarchy on kubuntu? Im using kubuntu 12.04 and libvirt 0.9.8.. Hi Daniel,
I’ve use cgroups on CentOS and I’ve noticed the cgroup hierarchy for libvirt that you mentioned. However, when I install all the packages for kubuntu, I noticed that the same structure does not get auto-mounted. I checked libvirtd.conf/qemu.conf and cgconfig.conf, all look same. Is there something i’m missing to get the same hierarchy on kubuntu? Im using kubuntu 12.04 and libvirt 0.9.8..

]]> By: Daniel Berrange http://berrange.com/posts/2009/12/03/using-cgroups-with-libvirt-and-lxckvm-guests-in-fedora-12/comment-page-1/#comment-25179 Daniel Berrange Wed, 30 Nov 2011 15:41:43 +0000 http://wordpress.berrange.com/?p=12#comment-25179 A task is *always* in exactly one cgroup. Init starts off in the root cgroup. New tasks that are started will be placed in the same cgroup as their parent. It is at administrator/application discretion to move tasks to non-root cgroups. A task is *always* in exactly one cgroup. Init starts off in the root cgroup. New tasks that are started will be placed in the same cgroup as their parent. It is at administrator/application discretion to move tasks to non-root cgroups.

]]> By: Philby http://berrange.com/posts/2009/12/03/using-cgroups-with-libvirt-and-lxckvm-guests-in-fedora-12/comment-page-1/#comment-25178 Philby Wed, 30 Nov 2011 14:56:50 +0000 http://wordpress.berrange.com/?p=12#comment-25178 Daniel, excellent article. A general question. With the following mount { cpu = /cgroup/cgroup0; cpuset = /cgroup/cgroup0; blkio = /cgroup/cgroup0; memory = /cgroup/cgroup0; } in cgconfig.conf, when I do a #cat cgroup/cgroup0/tasks 3 4 5 371 I can already see tasks in the cgroup even though I did not echo in the pid. How can I prevent tasks being added by default? Daniel, excellent article. A general question. With the following

mount {
cpu = /cgroup/cgroup0;
cpuset = /cgroup/cgroup0;
blkio = /cgroup/cgroup0;
memory = /cgroup/cgroup0;
}

in cgconfig.conf, when I do a
#cat cgroup/cgroup0/tasks
3
4
5
371

I can already see tasks in the cgroup even though I did not echo in the pid. How can I prevent tasks being added by default?

]]> By: Daniel Berrange http://berrange.com/posts/2009/12/03/using-cgroups-with-libvirt-and-lxckvm-guests-in-fedora-12/comment-page-1/#comment-20883 Daniel Berrange Fri, 15 Jul 2011 11:21:36 +0000 http://wordpress.berrange.com/?p=12#comment-20883 I'm not familiar enough with 'tc' to know what is going wrong for you there. There is, however, work taking place in libvirt to add in explicit support for integration with 'tc' for network I/O controls. We're not actually planning to use cgroups for this, since network controls really want to be per-VM network interfaces, not QEMU as a whole. I’m not familiar enough with ‘tc’ to know what is going wrong for you there. There is, however, work taking place in libvirt to add in explicit support for integration with ‘tc’ for network I/O controls. We’re not actually planning to use cgroups for this, since network controls really want to be per-VM network interfaces, not QEMU as a whole.

]]> By: Gianluca http://berrange.com/posts/2009/12/03/using-cgroups-with-libvirt-and-lxckvm-guests-in-fedora-12/comment-page-1/#comment-20252 Gianluca Mon, 04 Jul 2011 13:33:20 +0000 http://wordpress.berrange.com/?p=12#comment-20252 Question on Network I/O policy on guests Supose I have a rh el 6.1 or Fedora 15 host. Can I already use it for capping bandwith on guest? If so, what would be the device to use for tc comamnds? In my case I have: eth2 and eth3 that are part of bond0 vlan on bond0 that originates bond0.66 bridge on bond0.66 called brvlan66 guest that is on brvlan66 and, from a host point of view is using vnet3 device. Tried this: tc qdisc add dev vnet3 root handle 1: htb default 26 tc class add dev vnet3 parent 1: classid 1:11 htb rate 1024kbit tc class add dev vnet3 parent 1:11 classid 1:20 htb rate 1024kbit ceil 1024kbit prio 0 tc qdisc add dev vnet3 parent 1:20 handle 20: sfq perturb 10 Then if my vm has a pid of 2685 cgcreate -g net_cls:/w7x86lim cgset -r net_cls.classid=0x10014 w7x86lim echo 2685 > /cgroup/net_cls/w7x86lim/tasks But the network bandwith seems not limited... Also tried to put the into the same cgroup the "virtual cpu process" (correct?) that ha pid 2700, but no effect... Any hint of steps above and possible configuration to achieve network bandwith capping on rh el 6.1 and/or Fedora15? Question on Network I/O policy on guests
Supose I have a rh el 6.1 or Fedora 15 host.
Can I already use it for capping bandwith on guest?
If so, what would be the device to use for tc comamnds?
In my case I have:
eth2 and eth3 that are part of bond0
vlan on bond0 that originates bond0.66
bridge on bond0.66 called brvlan66
guest that is on brvlan66 and, from a host point of view is using vnet3 device.

Tried this:
tc qdisc add dev vnet3 root handle 1: htb default 26
tc class add dev vnet3 parent 1: classid 1:11 htb rate 1024kbit
tc class add dev vnet3 parent 1:11 classid 1:20 htb rate 1024kbit ceil 1024kbit prio 0
tc qdisc add dev vnet3 parent 1:20 handle 20: sfq perturb 10

Then if my vm has a pid of 2685
cgcreate -g net_cls:/w7x86lim
cgset -r net_cls.classid=0×10014 w7x86lim
echo 2685 > /cgroup/net_cls/w7x86lim/tasks

But the network bandwith seems not limited…
Also tried to put the into the same cgroup the “virtual cpu process” (correct?) that ha pid 2700, but no effect…

Any hint of steps above and possible configuration to achieve network bandwith capping on rh el 6.1 and/or Fedora15?

]]> By: Daniel Berrange http://berrange.com/posts/2009/12/03/using-cgroups-with-libvirt-and-lxckvm-guests-in-fedora-12/comment-page-1/#comment-10980 Daniel Berrange Mon, 21 Feb 2011 16:44:38 +0000 http://wordpress.berrange.com/?p=12#comment-10980 cgroups have been around since about 2.6.24 kernel, but not all the functionality was present at that time - stuff is added all the time. There's no cap or weight parameters to virsh schedinfo for KVM. There is only the 'cpu_shares' parameter, which is a tunable for the native Linux CFS scheduler. This is basically a relative prioritization metric. So if you have two guests, one with cpu_shares 1024 and the other with cpu_shares 4096, and both are contending for the CPU, one will get x4 the CPU runtime of the other. cgroups have been around since about 2.6.24 kernel, but not all the functionality was present at that time – stuff is added all the time. There’s no cap or weight parameters to virsh schedinfo for KVM. There is only the ‘cpu_shares’ parameter, which is a tunable for the native Linux CFS scheduler. This is basically a relative prioritization metric. So if you have two guests, one with cpu_shares 1024 and the other with cpu_shares 4096, and both are contending for the CPU, one will get x4 the CPU runtime of the other.

]]> By: Daniel Berrange http://berrange.com/posts/2009/12/03/using-cgroups-with-libvirt-and-lxckvm-guests-in-fedora-12/comment-page-1/#comment-10979 Daniel Berrange Mon, 21 Feb 2011 16:42:12 +0000 http://wordpress.berrange.com/?p=12#comment-10979 The "API" the kernel exposes for cgroups, is simply a virtual filesystem with a bunch of files you read/write to. There is a more formal userspace API provided by the libcgroup project that takes care of some of the crufty parts for you. The “API” the kernel exposes for cgroups, is simply a virtual filesystem with a bunch of files you read/write to. There is a more formal userspace API provided by the libcgroup project that takes care of some of the crufty parts for you.

]]> By: zvi dubitzky http://berrange.com/posts/2009/12/03/using-cgroups-with-libvirt-and-lxckvm-guests-in-fedora-12/comment-page-1/#comment-10561 zvi dubitzky Mon, 14 Feb 2011 16:47:48 +0000 http://wordpress.berrange.com/?p=12#comment-10561 1. starting at what Linux kernel version does the cgroup work ok ? 2. What is the meaning of the 'cap' parameter in the schedinfo virsh command (what are the units and range of values ) and what is the 'weigth' parameter .I did not find any documentation 1. starting at what Linux kernel version does the cgroup work ok ?

2. What is the meaning of the ‘cap’ parameter in the schedinfo virsh command (what are the units and range of values ) and what is the ‘weigth’ parameter .I did not find any documentation

]]> By: Luis A. Silva http://berrange.com/posts/2009/12/03/using-cgroups-with-libvirt-and-lxckvm-guests-in-fedora-12/comment-page-1/#comment-9336 Luis A. Silva Mon, 24 Jan 2011 22:23:32 +0000 http://wordpress.berrange.com/?p=12#comment-9336 API to develop an aplication on C/C++ with cgroups Hi All, I want to develop an application in C/C++ that allows to make calls directly to cgroups, do you can make Cgroups operations from a program in C/C++? if possible, there is an API to do this, wich packages i should install? or how i do this? or someone who can help me on this issue? Regards, Thanks adansilva05@hotmail.com API to develop an aplication on C/C++ with cgroups

Hi All,

I want to develop an application in C/C++ that allows to make calls directly to cgroups, do you can make Cgroups operations from a program in C/C++?

if possible, there is an API to do this, wich packages i should install? or how i do this?

or someone who can help me on this issue?

Regards,
Thanks
adansilva05@hotmail.com

]]> By: Daniel Berrange http://berrange.com/posts/2009/12/03/using-cgroups-with-libvirt-and-lxckvm-guests-in-fedora-12/comment-page-1/#comment-2537 Daniel Berrange Wed, 18 Aug 2010 14:26:08 +0000 http://wordpress.berrange.com/?p=12#comment-2537 There is no configuration required for the devices cgroup. libvirt automatically adds a 'deny all' rule for every guest. It then adds individual block devices that are associated with guest disks, along with a couple of common devices (/dev/null, etc) There is no configuration required for the devices cgroup. libvirt automatically adds a ‘deny all’ rule for every guest. It then adds individual block devices that are associated with guest disks, along with a couple of common devices (/dev/null, etc)

]]>