[asterisk-users] State of Asterisk+Virtualization+Timing

[Hans Witvliet]

On Tue, 2011-11-01 at 12:08 -0500, Tim Nelson wrote:
> Greetings-
> 
> I'm about to dive into the process of virtualizing some of my Asterisk (primarily 1.4.x) infrastructure. In the past, when looking at virt solutions, the primary issue preventing me from moving was the lack of proper timing. We do not need it for MeetMe but rather for IAX2 trunking. I'd like to use either OpenVZ or KVM, but each seem to have independent "issues" that need to be addressed:
> 
> OpenVZ - Better resource usage, lower overhead. Primary issue is how to grant access to host node timing source (physical device, or dahdi_dummy in /dev/dahdi/) to the containerized Asterisk process.
> 
> KVM - Higher overhead, easier installation, 'true virtualization'. Primary issue is not timing per se, but KVM scheduling. Timing source, while present from dahdi_dummy natively may still not get proper scheduling by KVM process. This could also affect general call quality (even non IAX2 trunked voice), DTMF, etc.
> 
> I have to believe there are others running virtualized Asterisk installations with some degree of success on OpenVZ or KVM. Care to share your thoughts?
> 
You mist out one more mature virtualization technique: XEN
Virtual machines can use  both hardware- or paravirtualization.
I have used both asterisk (1.4, 1.6.x and now 1.8) to separate machines
where people should do their sip-registration (internet / intranet /
pstn-gateway) and the actual dial-beast.

Main advantage for virtualization is (besides easy scaling) that you can
perform an upgrade in no-time (one VM-machine down, other up) Don't like
it: back in seconds!
Migration with an asterisk on real hardware takes much more resources.
Both in iron and in time.

hw