[asterisk-dev] IBM BladeServer / Clustered Linux

[Chris Lee]

Rich Adamson wrote:
>>>> We need the borg. We need each Asterisk box to be connected to the 
>>>> Hive mind such that all data that is accessible to one is 
>>> accessible 
>>>> in real-time to all.
>> Snip.....
>>
>>>> to a new box and have it pick up the processing of the RTP data and 
>>>> the voice frames on the next packet.
>>>>
>>>> That's a tall order!
>>> Yap... I entirely agree. It is too complex. Too many things 
>>> could go wrong.
>>>
>>  
>>> There is a better and much simpler way of solving this problem.
>> !!!!!!
>>
>>
>> Can you share here, or do you want 'Scoty' to beam me over for a quick
>> talk??? 
> 
> I'll jump in here with just one comment... think about the return on
> investment in real business terms for the high availability non-stop
> approach. Probably not justified in 99% of the cases.
> 
> Dropping the calls that are in flight is much closer to reality.

I am not an expert and know far to little to put this forward but the 
way I see it the following might work with only a few packets lost:
Distribute call setup info to all nodes in the cluster, so they all 
think they have the call set up (this needs modification so they don't 
actually do anything: no packets forwarded without packets arriving).
If a node goes down the management software tells another node to use 
ARP to take over the failed nodes IP details.
A few packets are lost in the interim but the call could maybe continue.