Every Volume inside a VM can be thought of as a linearly addressable set of fixed-size blocks. The storage of the actual blocks is orthogonal to the task of exposing a single block device for use by each VM. Bridging the gap between the VMs performing random access to Volumes and the storage of actual blocks is Archipelago: a custom storage handling layer which handled volumes as set of distinct blocks in the backend, a process we call volume composition. For the actual storage of blocks we are currently experimenting with RADOS, the distributed object store underlying the Ceph parallel filesystem, to solve the problem of reliable, fault-tolerant object storage through replication on multiple storage nodes. Archipelago itself is agnostic to the actual block storage backend.
Archipelago is under active development and will be available soon.