Terrence Nevins: We shipped three days early. Not a record here at Microsoft, but somewhat of a rarity in the world of software. It takes just a tad bit more effort beyond declaring code complete, test run completion and final signoff.
When there's a service involved you get a chance to do a whole heck of a lot more work before you get to call it done. It's roughly the same things customers will go through if they decide to host our server... with a few notable differences. We get to do it first. Which means we are documenting, double-checking and back-tracking at a constant rhythm. Some things go faster than you expected - and others go slower. We had about 16 machines to get ready for the pre-production and production system. Loading software for the basic platform which includes virus protection and monitoring solutions. Of course before this there was sizing the right capacity and ordering. And in conjunction with putting all these machines into service was the chore of configuring for remote access, naming the servers, associating workloads and about a dozen other pre-requisites.
Since this system was never before put into production we decided to take on an extra chunk of work and do some load testing on the PPE and production system. We we're pretty certain that our app would behave as good or better than our test lab runs. "At least on paper it should," we rationalized. But you can't ever be too sure. We wanted to validate our network settings and load-balancing. As it turned out, this was a good decision. We had to make some tweaks here, and some tweaks there but at the end of the day we were able to top the numbers we saw in test. This level of confidence would have been extremely hard to come by if we hadn't taken the challenge to push for some solid metrics.