[asterisk-users] Anyone using Asterisk on VirtualBox ?

[Jeff LaCoursiere]

On Thu, 1 Sep 2011, RSCL Mumbai wrote:

> 
> I tried and failed with VirtualBox too.  Timing seemed impossible to 
> maintain, even on beefy hardware (hexacore)
> with plenty of RAM (16G), and nothing else going on (single instance).  I 
> don't think VirtualBox is up to real-time
> stuff.
> 
> We use LXC now, and it is fantastic.
> 
> j
> 
> 
> Thx Jeff.
> 
> Kindly share some more details on the kind of hardware you are using, LXC 
> parameters and the kind of load the system can
> handle.
> 
> I am sure this will help me and more like myself.
> 
> Thx
> Sanjay
> 
> 
> My main interest of being on Virtual platform is portability / Backup.
> In case of any h/w issues, or crashes, simply copy the VM on to another box 
> and you are up in minutes.
> 
> 
> Sanjay
> 
>

Hi Sanjay,

LXC is more of a quasi-virtual platform - it doesn't give you hardware 
virtualization, but instead lets you share the kernel of the host between 
multiple instances.  To me this allows for multiple efficiencies and advantages 
that you don't get with hardware virtualization:

1) the host's memory is shared between all instances
2) the host's disk is shared between all instances
3) a shell on the host has access to the files in all of the instances

So an instance that is truly idle is taking up very little resource on the 
host.  Versus a traditional hardware virt, which even when idle has an 
appreciable chunk of RAM and CPU in use all the time.

For hosting lots of asterisk instances this is VERY efficient.

We have it setup such that the host runs an asterisk image that is the "PSTN 
gateway" and has dahdi loaded for timing and access to interface cards.  It 
accepts calls for subscribed DIDs and routes them to the appropriate instance.

Each instance has an asterisk process that is dedicated to a customer, which 
includes their own instance of FreePBX.  The dedicated asterisk instance uses a 
SIP peer connection to the asterisk running on the host which is its outbound 
access to the PSTN (or other instances).  The one gotcha I ran into was 
configuring the instance to allow access to the dahdi kernel module of the 
host, which is needed for timing for meetme (we still run 1.4).  The conf file 
needs to contain:

# dahdi
lxc.cgroup.devices.allow = c 196:0 rwm
lxc.cgroup.devices.allow = c 196:253 rwm
lxc.cgroup.devices.allow = c 196:254 rwm
lxc.cgroup.devices.allow = c 196:255 rwm

This is still in proof-of-concept mode for us, but we do have a half dozen 
customers representing about fifty seats running on it in beta.  No complaints 
in over two months, and the load average may as well be zero.

The machine is a quad core Xeon (X3450 @ 2.66Ghz) with 8G RAM, running Ubuntu 
11.04.

Each instance is a subtree of the host's filesystem, by default (at least in 
Ubuntu) under /var/lib/lxc.  We created a template with a full asterisk and 
FreePBX installation.  To create a new instance we simply untar the template 
and run a sed script over a set of files to give it an IP address, hostname, 
and minor edits to various asterisk config files.  I haven't done it yet, but I 
intend to create a mirror of the host machine on another box with rsync, which 
will serve as the backup.  At some point I would like to have the instances 
running on both mirrors with failover.

LXC docs basically suck.  If you do go down this road, you will have to be 
prepared to glean as much as possible from notes various people have posted.  I 
settled on Ubuntu 11.04 as a base because a lot of LXC specific scripts have 
been created to help with management.  Even so its kind of flaky shutting down 
and rebooting the instances.  Once they are running as you like it is stable, 
but I had a lot of weird things happen along the way as I was tweaking.

OpenVZ is the older and more mature equivalent, and may be a better choice to 
start, but it is not built into the kernel as LXC is.  I don't have an real 
comparisons to provide operationally, but I can vouch for LXC being stable 
enough for production use so far.  I haven't stress tested it yet to see how 
many instances we can provide on a single host, but am hoping it to be a 
function of the number of simultaneous calls rather than the number of 
instances...

Would love to hear from anyone else that is using LXC, especially in 
production.

Cheers,

j