I've always built that rather complex infrastructure on top of (ie, using) processes and channels.
It's "complex" not just because you might want different storage methods (eg, file system, or this or that database),
but providing fault-tolerance through replication on top of that (say) is much easier built using Loop and its abstractions than
having it in the run-time system. The algorithms for fault-tolerance using (say) the replicated state machine approach
are not huge, but they are non-trivial and relatively intricate.