[asterisk-users] Managing complex setups with Asterisk

[Johan Wilfer]

2012-11-08 00:26, Jeff LaCoursiere skrev:
> On 11/07/2012 05:20 PM, Jeff LaCoursiere wrote:
>> On 11/07/2012 02:16 PM, Johan Wilfer wrote:
>>> 2012-11-07 20:49, Jeff LaCoursiere skrev:
>>>> Just to chime in, if you REALLY want multi-tenant, it is super easy and
>>>> surprisingly efficient to use kernel level virtualization to run
>>>> multiple instances of asterisk (and even FreePBX).  We use LXC to do
>>>> this.  The "host" runs an instance that has the dahdi hardware,
>>>> drivers,
>>>> and upstream connections.  The "clients" have SIP connections to the
>>>> host for all inbound/outbound, so you have a central place to
>>>> collect/process CDR records for billing.  Getting your phones to
>>>> connect
>>>> to each instance is an exercise for the network admin ;)
>>> Any quirks / observations you have running LXC? We run OpenVZ now with
>>> the same setup and it works very well. But as Debian will not support
>>> OpenVZ in the next version we are looking for alternate solutions..
>>>
>>> Do you run Dahdi run Dahdi for timing / meetme on both the "host" (HN)
>>> and the "clients" (VE)?
>>>
>>> Distribution?
>>>
>>> Any other pitfalls or recommendations with LXC?
>>>
>>>
>>
>> Since moving to Ubuntu 12.04 server, LXC mgmt has been much simpler
>> and stable.  Had some troubles with Ubuntu 10, though that was our
>> proof-of-concept, and mainly just with getting a template finalized.
>> Shutting down a container, back then, was a scary and often fatal
>> thing to do.  WIth 12.04 I have had zero LXC related issues in roughly
>> six months.  Have a few dozen companies running on the platform and
>> getting ready to white label the infrastructure to several resellers.
>> In our lab we have managed to get 200 instances, with FreePBX, running
>> simultaneously (though idle) on one host. Each (optimized!) container
>> seems to eat about 75M of RAM.  Our latest tweak is to make all of the
>> containers internally addressed on an OpenVPN-only accessible virtual
>> LAN, and are only distributing telephony hardware that can connect to
>> the platform natively (still on a search for the right ATA, though
>> getting by with DD-WRT router in front of Cisco ATA).
>>
>> Cheers,
>>
>> j
>>
>>
>
> Realized I didn't answer your questions.  Yes, the host runs Dahdi for
> timing, entirely for Meetme in the containers (host is just for
> transport), and the driver is exposed to the containers in the LXC conf
> file.
>
> LXC does NOT have the same level of operational controls over the
> containers, yet, as OpenVZ (like limiting resources).  That hasn't
> really been an issue for us so far.
>
> Cheers,
>
> j

That's very cool.
I've also looked at Ubuntu server for their long support-time. I 
definitely have to test this setup.

Thanks for the insight!

-- 
Johan Wilfer

JT Technologies & Telecommunications AB
Jabber: johan at jttech.se | Web: www.jttech.se