"We install the product, get the users trained, migrate our data and are up and running like clockwork," fish says. "Then the odd things start to happen.

"The application has sudden little freeze-ups and drop-offs on the network. Just the one app -- everything else on the PCs works just fine. And it's not all at once: one here, two over there, but never a whole bunch at one point in time to indicate what it is that's causing the problem."

Fish and his cohorts start troubleshooting. They rebuild PCs and servers from the ground up. They strip out all unnecessary software. Still the hiccups continue.

Clearly, it's the new application. The new app's data is being transmitted via TCP/IP, and none of the other IP-based software is having problems. And the new app uses its own proprietary IP stack instead of a standard model. That, fish figures, is the likely culprit.

So it is that fish finds himself sitting in on a conference call with the software vendor. Nearly two-dozen IT staffers are in the conference room, and just as many are on the call on the other side.