Scaling high computation load wrt. the middleware
Four paradigms: - Map-Reduce/offline/batch for latency insensitive cases - Shared caches for low-latency computations + naturally partitionable data - Tuplespaces/Grid movable code (at least for JavaSpaces or LISP-based solutions), customizable logic etc. - Traditional messaging systems for ordering-sensitive data, coordination - Start hacking your own (Cassandra?), custom aggregation trees
- Use a wiki (padded room option)
Group G1, Dan and Kjetil