We explore the implementation details and consequences of running a distributed file system in the cloudlet. Using a centralized Samba server VM that exports a Coda mount point, we allow client VMs within the cloudlet to have a common shared file system. We use the tenancy system in the cloudlet for access control, meaning that only VMs owned by the same user or organization share the file system mount.
Along with the discussion of the implementation of the system, we investigate the effects of our architecture on the cloudlet ecosystem. Particularly, we focus on its integration with the cloudlet’s VM migration feature. Running a centralized distributed file system per tenancy allows the file system servers to communicate with each other across cloudlets for prefetching highly used files in parallel with VM migration. Furthermore, we argue that our architecture integrates well with existing software because it provides UNIX semantics and we show we can run Hadoop, a widely used distributed data analysis framework, on top of it.