> With Spotify, the aim was low latency. If you click on a song, it should start playing immediately.
That was not their aim at all. Their aim was lowering bandwidth costs by shifting the content delivery costs over to the end users. This was made clear by the fact that when you played a song, it started downloading the song from AWS S3 immediately to minimize latency, then once the P2P sessions fired up, shifted over to that stream instead of their costly S3 one.
I had always assumed that the P2P was to help mitigate the cost of data so that the Spotify team could better use their limited (at the time) funds to secure more music contracts. Were operational costs not as drastic as I might be assuming?
Back when the only data center was in Stockholm, the only way to meet latency requirements was P2P. Now that we're a grown-up company we've disabled that and use CDN's instead.