This is open-source version of CPFS, the ClusterTech Parallel File System. CPFS is a high-performance distributed parallel filesystem which is fault tolerant while being very easy to set up. The interface is highly POSIX compliant, so most applications do not need to be rewritten to enjoy the high speed and fault tolerance provided by CPFS.
A CPFS cluster consists of 1 or 2 (preferred) meta-data server, and a multiple of 5 data servers (compile time constant for now). Meta-data servers store directory data and replicate them like a RAID-1 disk array. Data servers store file data and replicate them like a RAID-5 disk array. The server data is accessed via FUSE-based clients. Node-local cache provided by FUSE is used, with cache coherency implemented. Meta-data is fully coherent. Close-to-open coherency is implemented for data.
The software supports automatic failover and data resynchronization upon server crash and recovery. At present, the system is not responsive during resynchronization. (We are now trying to modify the software to allow partially online resynchronization.) Unlike an actual disk array, the replication is file-level, allowing fast system recovery if the amount of data to resync is not large.
The project is hosted in GitHub. In its "releases" section you can find installation packages for it. Further information can be found as follows:
CPFS is released under AGPL 3.0. It uses a few third-party libraries, with their respective license.