I wish Zooomr the best of luck launching Mark III. At the same time, as a developer interested in their open API, I can’t help but wonder about the cause of the complete failure of Mark III to launch anywhere near on time. It’s now six days late (not counting the aborted attempt in April) and has been down continuously for a week. That’s gotta hurt. Mark III (that’s Zooomr version 3, by the way, not a torpedo designation) claims to have been 9 months in development and has 250 new features. It’s also been hinted at by Zooomr that they are migrating at least part of their infrastructure away from Amazon S3 for performance reasons.

I hope that Zooomr elaborates on their launch issues on their blog (after Kristopher gets some much needed sleep). Until they do, I can only speculate. But one thing that comes to mind immediately is the craziness of bottling up 250 new features and deploying them all at once. Rolling out one new feature is difficult enough—multiplying that by 250 is extremely risky. It’s a very Web 1.0 way of doing deployments. I can’t say why they did it this way but it’s obviously not going well. And what role does S3 have to play in this, if any? Is it taking longer to migrate data out of S3 into a local storage system than was estimated? If you’ve got terabytes of data invested in S3, how do you get it back out again? Why didn’t they start the data migration prior to taking the site down and then transfer the balance afterwards?

When it comes down to it, I’m sure Zooomr’s problems aren’t caused by one thing that anyone can point to and say: “Look! It was this thing here.” Deployments always involve some level of risk. And I understand the circumstances that can sometimes lead to the need for large, monolithic deployments. But doing a storage system migration and adding 250 new features all at once seems downright irresponsible. Scoble is right on the money when he says you need a parallel set of servers to play around with. The thing is, they’ve presumably already got the infrastructure they need to run Mark III. So why not segment that to keep the old site running while they brought Mark III online?

As I write this, the Zooomr site shows a new launch ETA of 7pm PDT (just a few minutes from now). I’ll raise a cold one for you, guys.

Comments

1. Norby on 2007-05-30 21:57:05 wrote: I think they had split them up at one point, but were unable to sustain the live traffic load while continuing development. Which is, kinda too bad, as I hate to set anything live that hasn’t been thoroughly tested and without a fallback plan to boot… -//