I had the opportunity to lunch with Paul Patrick, BEA's chief architect for AquaLogic, yesterday in New York. Paul provided some interesting food for thought, which I intend to explore in future editions of this Weblog.
Namely, that service-oriented architecture (and yes, that includes ESBs) is developing within organizations the same way the Internet and World Wide Web grew -- as "neighborhoods" of systems that spring up to serve more localized requirements. Eventually, just as was the case with the Internet, these neighborhoods will start joining into a greater federation.
Paul, a veteran of CORBA who also played a role in DCOM's proliferation in enterprises, also carries another message -- there's nothing truly new or revolutionary about SOA; we've been there, done that before. What has tripped up previous attempts at reusable Lego-block architectures is that they were IT-centric, and didn't directly address business problems.
Layer 7's Scott Morrison, writing in ComputerWorld, echoes Paul's sentiments. Morrison observes that "anyone who lingers long enough in computing begins to develop an overwhelming feeling of deja vu," and SOA is bringing on such feelings. Performance issues, for example, are rearing their heads again.
We can learn from history, however -- particularly with the way the Web developed in the 1990s, Morrison says. "When cryptography overhead became a limiting factor in Web performance, it was moved to dedicated hardware-based Secure Sockets Layer (SSL) accelerators. When applications outgrew single servers, they moved to clusters with dedicated, hardware load balancers distributing the volume. When authentication and authorization became too expensive to manage locally on the application server, it moved to authenticating reverse proxies."
Morrison puts it this way: "History repeats, and performance is now a challenge for SOA. Today's issues are more complex, but the Web offers lessons in how we might approach these problems."
And, as Paul Patrick puts it, the early days of the Web are providing a lesson in how we can expect to see SOA evolve -- and hopefully, learn from earlier attempts.