Rourke on SOA

Three articles discussing service virtualization have popped up in the last 24 hours. Not all of them use the term service virtualization, but they’re all referring to the concept.

Todd Biske posted an article on his blog that eloquently explains the need to scale services differently from web applications — the load characteristics are not the same. Todd refers to BEA WebLogic Server Virtual Edition as a tool that can help quickly bring up extra instances of services hosted in a traditional application server to meet load spikes, but ActiveMatrix Service Grid is an even better option. Last week TIBCO’s Matt Quinn talked about service virtualization at a Gartner event and specifically addressed the value in scaling services at the service level, rather than scaling an entire application server instance. ActiveMatrix Service Grid is the only product on the market today that allows you to do that.

Todd finishes his article by comparing SOA to a “virtual” or “distributed” mainframe. The analogy is interesting, because service virtualization eliminates quite a bit of the complexity that comes from smearing applications out across services, containers, languages, and geographies. On a mainframe you could invoke any functionality by making a function or procedure call. With ActiveMatrix you have ability to do that once again, even if the service you’re invoking is on the other side of the world, running on a different operating system, written in a different programming language.

On a mainframe you can control how much processing power goes to each component without having to deploy extra instances of those components on other machines; this gives you the advantage of co-location even while scaling the power given to that component. ActiveMatrix allows you to deploy many instances of a highly loaded service to scale that service, but you can deploy some of those instances co-located with the other services it needs to communicate with the most. In this way you get the in-proc collocation while still getting the scaling flexibility.

Of course, mainframes aren’t the desired end state. There are plenty of good reasons we moved away from mainframes, but if we can get back some of what we lost by making that move, all the better. Everyone wants to have their cake and eat it too.

Scaling aside, the abstraction that you get from virtualization appears to be the main reason Delta chose ActiveMatrix. Joe McKendrick mentions this in his SOA in Action blog. Joe also refers to Todd’s article but doesn’t specifically mention that you can solve both the scaling and abstraction problems with service virtualization.