Thanks to this change: - all the maps bucket -> buffer are indexed by int's instead of string. No need to do: byte[] -> string -> hash conversion on each access. - buckets are strongly typed in backend/mvcc API.