Autonomous Services and mission critical requirements

Our question this week comes from Eli who asks:

"Hi Udi,

Thank you for you insightful comments on SOA.

We are re-designing an existing distributed system, but now we want to make it based on SOA and messaging. And we want to make it simple :)

I really like the autonomous service and its scalability, and the messaging approach. It helped me see exactly what I want / need from WCF.

But, if we need a failover mechanism for a service, in a mission critical environment, then the model seems to break. IMHO, we need another layer / mechanism to route a message between identical autonomous services.

Can you comment on the failover / mission critical issue?"

Well, Eli, happy to be of service.

My experience has taught me that knowing what you want is half the solution. In the scenario you describe (which is really quite common), you’d want clients of the service not to be aware of the failover, while allowing additional servers to take the load. You could configure various messaging products to give you just such a solution – each being somewhat different from the other.

Using MSMQ in a cluster could give you the first part – clients continuing to work with the same virtual server. On the flip side, for scale out, because MSMQ doesn’t support remote transactional reads yet (will in the LongHorn timeframe) you’d need some kind of dispatcher to spread the load among available servers. However, all this is just infrastructure implementation – it doesn’t make any difference to the architecture or design of the system. Of course, at deployment time you’d want to have everything configured for optimal performance.

I hope that answers your question.

--

If you happen to have any issues with the architecture or design of your service-oriented solution, don't hesitate to send me an email to Blog@UdiDahan.com.

Posted by Udi Dahan at 03:40 PM  Permalink