pNFS Block Server Support Is Coming To Linux 4.0 (3.20)

[phoronix]

Phoronix: pNFS Block Server Support Is Coming To Linux 4.0 (3.20)

The nfsd changes for the Linux 4.0/3.20 kernel provide pNFS block server support...

http://www.phoronix.com/scan.php?pag...0-Block-Server

[xeekei]

Maybe XFS will get a bit of a boost development-wise now when RHEL 7 uses it as default.

[q2dg]

This means pNFS will be able to be a replacement of NBD??

I ask for choosing the best in the future.
Thanks.

[cb88]

I ask for choosing the best in the future.
Thanks.

Not really NBD is more like iSCSI. Only one user can access the disk.... I have used NBD in the past for mounting large storage on a small machine IE a router or vintage computer (Sparcstation LX in my case).

NFS and hence pNFS are network file systems ... and can be accessed by multiple users at the same time. Its a very cool development though.

[dnebdal]

Not really NBD is more like iSCSI. Only one user can access the disk.... I have used NBD in the past for mounting large storage on a small machine IE a router or vintage computer (Sparcstation LX in my case).

NFS and hence pNFS are network file systems ... and can be accessed by multiple users at the same time. Its a very cool development though.

On the flipside, won't this limit you to clients that all understand the same filesystem? I've got a FreeBSD/Linux mix here, with some samba serving of the same files to windows users ... and I have a feeling I'm not the primary target for pNFS.

[tessio]

It's comparable with ceph?

[dnebdal]

It's comparable with ceph?

Or Venti from plan9, or iSCSI - it's a block storage that requires the client OS to do something intelligent with the blocks (like, say, putting a file system on them). Which is perfectly nice and useful, though it's almost required that your clients all run the same OS.

Of course, you could set up a file server that served the "difficult" clients from files stored in pNFS/ceph/whatever - blocks, I'm just not sure if it would be worth the extra management compared to just serving files directly from a RAID. It looks like Ceph could get you some extra redundancy in a way that's harder to do with fileservers, so that might be useful in certain cases?

[pgoetz]

I'm confused that this can be a feature. Once you start letting clients access the datastore as a block level device, aren't you basically throwing all the features of NFS out the window and getting the equivalent of iSCSI?

[pgoetz]

I'm confused that this can be a feature. Once you start letting clients access the datastore as a block level device, aren't you basically throwing all the features of NFS out the window and getting the equivalent of iSCSI?

Actually, never mind; I was confused. The whole point of pNFS is [to provide client's with metabacked block level access to storage. The actual question is how did linux have any pNFS support prior to this?