These two boxes will serve nfs filesystems for dumps generation; they will be mounted only on the snapshot hosts. Files must be kept for at least two runs (current run, previous full run); new files must be copied over to dataset host after generation, so they can be rsynced elsewhere and served to the public. The dump monitoring service expects live status information; this will have to be changed.
Architecture for the above needs to be worked out, then implemented in puppet manifests and deployed.