So I'm somewhat familiar with NFS and ilk, and I realize that NFS 4 is adding a host of new features, when I stumble onto pNFS. It appears to be "parallel" NFS storage, but the diagrams and explanations that I'm getting are about as clear as looking through a greasy Pyrex dish. It also appears that it will "go live" sometime in 2010.
So, I was wondering the following questions:
- Just what does the "parallel" stand for? Parallel I/O? Access? Parallelism? There are several different connotations.
- Is it similar in concept to OpenAFS file services or Windows DFS?
- Does it provide redundant storage? (i.e. spreads out copies of data between servers)?
- There is mention of an object store. Is there a new API that will make this available? Or will this just use existing filesystem semantics to read/write objects?
- There is mention of block storage. Will this allow me to cluster a drive with other clients (i.e. shared filesystem)? And does this imply that I can have my block device on server A servicing multiple clients B and C via pNFS? While you can share a physical disk with iSCSI, this would certainly provide the "missing layer" (filesystem)...
- It appears to use a metadata server similar to CXFS, i.e. there is a separately maintained metadata server that determines what client receives access to what data. Is this restricted to a single metadata server, or can I have some redundancy and load-balancing here as well with multiple servers?
Parallel everything, it seems, and you can do redundant storage with it.
Shared file system near as I can tell.
The metadata server can be clustered, and that can be HA, with redundancy from many cheap disk servers.
Or, just google it. pnfs, there is plenty of info.
It does not have much of the interesting bits of AFS.