That has been discussed in this topic:
My conclusion:
- You will probably need a local file system with a copy of data (or at least of your “hot” data).
- There are various approaches and mechanisms that will make certain workloads (including low-intensity write workloads) suitable for keeping a single copy on the Project Safe Network.
I don’t think it’s bad, but if you lived under the impression that you’ll have a global file system with distributed locking running over an extremely loosely coupled network with tens of thousands of clients dropping out and joining the cluster every hour, well, then you had unrealistic expectations.