Amazon S3 changes the game
Foneshow has a lot of transcoded audio content - shows that have been called in, podcasts that we've ripped to WAV - that we want to keep available on a few seconds' notice in case a user asks for them, but that aren't in any user's playlist at the moment and don't need to be on the telecom server's hard drives. At present, about 90% of our stored content is in this "waiting for the long tail" category; that'll go down as more users subscribe to a greater variety of shows, then up again as we archive shows over time. It's an important piece of the business model - but it's an expensive asset to store, and a substantial part of our budgeted running costs have been based on the assumption that we'd soon need a couple of terabytes of colocated storage.
Until, over the weekend, I copied all the archival audio files up to Amazon's Simple Storage Service, and taught the Foneshow application to copy them back as they're requested. Now, every night, we sweep all shows that aren't in anyone's playlist up to S3, and we can delete the local copies whenever we need space. The local disk becomes a cache for the main S3 data store, and instead of buying big disks now in hopes of needing them later, we pay only for the space and bandwidth we use.
Total cost to date? Nine dollars and fifteen cents.
This is absolutely transformative for pre-capital startups. That first investment, just to get the hardware you need to test the business idea, can be deferred until after the idea is proven. And done right, there's very little added risk; if S3 proves unstable, or Amazon starts messing with the terms, we can just scale up the local storage, stop deleting old files, and the application will forget all about S3.
I hope that doesn't happen, though, because Amazon has another piece of wild magic in beta: their Elastic Compute Cloud. More on that later...