Everything looks good on paper. When you're planning out a major new IT initiative, say a service-oriented architecture, you never create a space on the schematic which says "you will be plagued by consistent performance problems."
No one plans for that. Of course it could be argued that by not taking those potential issues into account during the conceptual phase you are, in a sense, planning for performance headaches. For instance, do you have a low latency strategy inside your SOA plan? When you're dealing with broadly distributed Web services and the verbosity of XML, high latency can rise up and bite you right on the butt. The best conceptual Web service on the planet can fall flat if it can't perform and scale.
Have you considered how multi-core processing can aid your SOA efforts? Hardware has become a big part of the SOA equation. The IT managers have a stake in this as well. They aren't going to get behind SOA projects if it means their lives become more difficult.
This week we at SearchWebServices.com will be looking into one major potential SOA bottleneck, data performance. At the end of the day, every application relies on data. If you've got an architectural plan for how to represent data, how to abstract it, how to share it, how to migrate it, then it's going to make your applications clunky. As it turns out, researchers are discovering that IT shops are waking up to the need for data architecture as part of their larger app dev plans.
It's no small part of the reason why Red Hat bought MetaMatrix last week. Data integration could be the big thing in the SOA space in 2007. It was governance in 2006 and data issues are simply too important to ignore. SOA has moved past the trial stage in many companies and performance is becoming a critical piece of the equation.