[Freeswitch-users] Advice on scalable design pattern

[Cal Leeming [Simplicity Media Ltd]]

Hello all,

I'm currently looking at the various different ways you can deploy
FreeSWITCH in a scalable manner, but struggle a little bit on design.

The sweet spot I'm trying to find is one where I can scale out capacity by
simply throwing more servers at it.

In an ideal world, this would mean support for;

* Have multiple users from multiple domains to be spread over multiple
servers... a single domain should not be restricted to a single FreeSWITCH
instance
* Have no single point of failure within the structure
* Have no single point of bottleneck within the structure
* Should not use OpenSIPS.. (I suspect this might get me a lot of flack,
but seriously, I'd rather write my own in python or ZXTM traffic script
than use OpenSIPS lol).

So far, the best option I can come up with is (although I'm not sure if
it's the best available);

* Proxy sitting in front of all backend FreeSWITCH instances, acting in a
media proxy fashion only (dual pair of proxies in active/passive mode)
* Proxy tracks registrations to the appropriate backend instance, and makes
their session sticky
* If backend instance needs to make a call to another user in the same
domain, it bridges to the call to back to the proxy, the proxy then
determines which other FreeSWITCH instance has the user then routes the
request accordingly. If the call is to an external destination, the proxy
will route it to the traffic aggregation switches (which is basically
another pair of FreeSWITCH instances), which then gets routed to the
upstream provider.. this means you only have to maintain 2 sets of trunk
configuration.. so when you need to scale out your freeswitch backends, it
doesn't require putting in a request to your upstream providers for an
additional set of trunks.
* The bottleneck within these clusters is the dual proxies in
active/passive mode.. you could fix this by allocating customers to a
specific cluster (rather than instance), thus controlling which customers
go to which proxy.. if an entire cluster dies, you can re-route that
clusters traffic to a different cluster.

The other simpler option is to allocate domains to a specific backend
instance.. but this really doesn't feel clean.. it means a customer cannot
scale past the floor limit of a single instance, it has less redundancy,
and overall just feels wrong.

Any general thoughts/comments on this would be much appreciated.

Thanks

Cal
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.freeswitch.org/pipermail/freeswitch-users/attachments/20130314/c44fab27/attachment.html