You know that a standard works when you can go from one implementation to another, without too much effort. BPEL has been promising such interoperability for quite some time, but to the extent of my knowledge, it has never been demonstrated at a large scale in a production environment. Until now. Over the weekend, Coghead went from one BPEL engine to another, and it worked without a glitch. Today, we can safely say that as an industry standard, BPEL really works.

Coghead has developed one of the most innovative Web 2.0 development tools, architected on top of a SQL database and a BPEL engine. For the later, they originally started with the product of one of our competitors, then decided to migrate to Intalio|Server for performance reasons. One of the performance issues they were facing was that they had to manage very many different process models—up to 100,000 on a single server. And with so many models to manage, re-starting the server was taking longer and longer—up to 6 hours for 20,000 process models. Eventually, it became unmanageable, and they started working with Intalio.

We initially did some benchmarking, then made some modifications to our engine in order to support the lazy loading of process models, and finally prepared everything for a migration that would take the processes of thousands of users from one engine to another. We completed the migration over the week-end, and it worked. Restarting the server now takes 7 seconds, down from 6 hours…

In their blog, the good folks at Coghead described this exercise as brain surgery. Being a pilot myself, I would compare it to changing the engine of an airplane in mid-flight. However you look at it though, this success story demonstrates that BPEL is ready for prime time, can be used to build very large process systems, and is available from multiple vendors with near-perfect interoperability. Finally…

Congratulations to the Coghead team and to Matthieu for an amazing achievement!