It keeps peer data exclusively in memcache -- would make it difficult for Google to comply with a subpoena, and foster maximum plausible deniability for all parties.
Amazon provides the tracker, but you do have to keep a copy of the files you are providing on S3, increasing your liability if you are doing something illegal.
App Engine gives you 1.3M free requests per day. Assuming that each peer announces and scrapes every 10 minutes, that's over 4,500 simultaneous peers. You get 6.5 CPU-hours per day, which leaves you 18 CPU-ms per request.
That is what I'm trying to figure out. Atrack is designed to make as efficient use of the quota as possible, and so far so good, but I hope all this publicity helps a bit with the stress testing.
It keeps peer data exclusively in memcache -- would make it difficult for Google to comply with a subpoena, and foster maximum plausible deniability for all parties.