I'm teaching OO design and UML again this term, and one of the things I emphasize to the class is the dangers of coupling. (Get your mind out of the gutter, I mean classes with dependencies on each other.) It's not about calling methods of each other necessarily; it's more about if-I-change-this-one, I'll-have-to-change-this-one-too. Changes that ripple through a system are expensive and dangerous.

When I am helping clients with interop, they are often surprised to learn how entire applications and libraries can depend on each other without ever calling each other's code. For example, App A writes a record to a database table. Service B checks the table regularly for new records (or records with a 0 in the Handled column, or whatever) and calls a web service (or whatever.) Those applications are now coupled - if a change in one necessitates a change in the format of that table (or its name, etc) then the other must be changed too. Thinking ahead and doing all you can to reduce this kind of coupling is part of the challenge of doing good application integration, even if there are no interop calls in the solution.

And then there's performance. So often left until last, it provides another consideration that you should ideally have in mind all the way along. And as Rico Mariani points out, it also couples applications and libraries you may have thought were independent:

Two subsystems that both (loosely) use 2/3 of the L2 cache are going to use 4/3 of a cache... that’s not good. There may be no lines between them in the architecture diagram but they are going to destroy each others ability to work.

Sound advice as always. Please read the post, and keep one more "don't forget" floating around as you design and architect your systems and solutions.

Kate