should be configurable as to using plain replication and erasure coding to distribute the files over the cluster. we'll start with plain replication since it's much easier, but long term, it would be nice to erasure code and distribute chunks.