[asterisk-dev] IBM BladeServer / Clustered Linux

Greg Boehnlein damin at nacs.net
Mon Feb 13 13:33:31 MST 2006

On Mon, 13 Feb 2006, Patrick wrote:

> On Mon, 2006-02-13 at 14:54 -0500, Robert Bedell wrote:
> > I remember a while back there was an article about IBM using their
> > bladeserver architecture to implement a clustered asterisk.
> > Specifically, they had multiple asterisk blades, and when one went
> > down the call would be picked up by another blade without being
> > disconnected. 
> > 
> > I post this to the –dev list because I’m interested in the technical
> > details of what they did (those I can gather without an NDA, anyway).
> > Anyone have an idea on this? 
> > 
> > If you know the name of the people at IBM who participated in the
> > development it may help as well.  Sorry for any noise ;(
> A few months ago I asked the guy at IBM working on this project. He did
> not give me any information. Said it was still in a research phase and
> not even remotely close to a real product. Maybe things have changed in
> the meantime but I have not seen any other information on this project
> floating around.
> Regards,
> Patrick

I believe that this is based on the presentation that was done in Anaheim 
at Astricon 2005. As I understood it from being in the audience, IBM was 
studying the requirements of what would be needed to do a real-time 
clustered VoIP implementation w/ Asterisk whereby the failure of a single 
server in the cluster would not drop the calls for the users on that 
server. This involved some major changes to virtually every aspect of the 
software such that VoIP frames and state information would have to be 
distributed in realtime accross the nodes such that any node could pick up 
the call IN PROGRESS from the failed server.

This was enough to make me realize that it is highly unlikely that we will 
EVER get to that sort of level with Asterisk. I.E. a fully fault-tolerant 
PBX and Switching Core that runs on multiple boxes is most likely a huge 
overhaul of the core.

However, there is no reason that RealTime cannot be extended to provide 
High Avilability and Fail-Over capabilities between multiple boxes, albeit 
at the expense of losing the calls on a server if one of the nodes bites 
the dust.

    Vice President of N2Net, a New Age Consulting Service, Inc. Company
         http://www.n2net.net Where everything clicks into place!

More information about the asterisk-dev mailing list