[Serdev] A suggestion for how SER should focus - was: So who/what is SER for, anyway?

[Greger V. Teigre]

Well, how often does a mobile call fail anyway? So much for 5 nines...
What's the cost of transparently failing a transaction over to another 
server?
g-)
PS. I agree with the general idea that auto-failover is good. I'm not 
just convinced that we are not fooling ourselves. IMHO, due to the cost 
of failover, it should seen as optimization: "Pre-mature optimization is 
the root of all evil."

Martin Hoffmann wrote:
> Greger V. Teigre wrote:
>   
>>  
>> "replicating its transaction state", you mean a SIP transaction?  If 
>> that's what you mean, this is an example of why telco-apps get so 
>> expensive. What if the server goes down in the middle of the 
>> transaction?  It fails, hey, that's a bummer, so it has to be resent, by 
>> the UAS. Why would you spent an immense amount of money to make sure 
>> that the transaction can fail over?! 
>>     
>
> INVITE transactions can go on for a very long time (technically,
> forever) and, what's more important, their progress is user visible. It
> actually does make sense to continue them on a stand-by box if the
> primary box fails.[0] For everything else resent is good enough.
>
> Regards,
> Martin
>
> [0] IMHO, this is a flaw in SIP. Instead of one very long transaction
>     with a three-way handshake and related extra transactions on the
>     side (PRACK), a series of short transactions would have been better.
>
>
>   
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.iptel.org/pipermail/serdev/attachments/20070130/92ef1caa/attachment.html