In-Kernel SMB3 File Server Looks To Land In Linux 5.15

[jacob]

Two things are bad with NFSv4x:

- crappy documentation
- user mapping is cr*p. Don't event try to use it, at least not without KRB5 ( annd stay away from KRB5!).

Which still leaves some space for it in small companies with only a couple machines or manageable clusters, within which one can keep UID/GIDs in sync.
Once that is done, it works well.

But what would be the point of doing that? SMB just works, and is also supported by Windows (obviously) and MacOS.

[Brane215]

But what would be the point of doing that? SMB just works, and is also supported by Windows (obviously) and MacOS.

NFSv4 can be way faster than SMB. it is also much closer to the metal and thus native disk access. So one can use various stuff that intimately depends on some filesystem features and it will work fine. Like git and similar tools. It can also handle SELinux context attributes. Also, you can run pNFS on it, if underlying filesystem on the disk allows it ( XFS) etc.
NFSv4 is much closer to Linux infrastructure, not just because it runs most of the stuff in kernel, for both server and client.

[Brane215]

I'm not aware of any Realtek-based NIC implemented RDMA.

If budget is a concern your best option is to go with used Mellanox InfiniBand cards or, not RDMA but iWarp, used Chelsio cards

BTW, AFAIK most new 2.5G NICs do offer RDMA, which might be a good reason to get them, even if I run them at just 1Gb/s over old switch...

[jacob]

NFSv4 can be way faster than SMB. it is also much closer to the metal and thus native disk access. So one can use various stuff that intimately depends on some filesystem features and it will work fine. Like git and similar tools. It can also handle SELinux context attributes. Also, you can run pNFS on it, if underlying filesystem on the disk allows it ( XFS) etc.
NFSv4 is much closer to Linux infrastructure, not just because it runs most of the stuff in kernel, for both server and client.

In my opinion though these are non-problems. NFS (including NFSv4) was created with the diskless workstation in mind, which is the exact opposite of today's usage scenarios. It may be theoretically possible to use git etc. over NFS but I wouldn't wish it to my worst enemy, and there is absolutely no reason to do it since nowadays even the cheapest laptop comes with half a terabyte or so of local storage. Shared filesystems are now used much more like a kind of FTP, where you download your stuff and then upload it again once you are done working on it - something that can (and should?) be replaced with webdav or sftp, rather than a pseudo-transparent network mount.

Besides NFS implementation in Linux has never been great and AFAIK Linux doesn't even support NFSv4 ACLs.

[Brane215]

In my opinion though these are non-problems. NFS (including NFSv4) was created with the diskless workstation in mind, which is the exact opposite of today's usage scenarios.

Not true. I do most of my work over NFS4. I'm not there yet with booting over it, but I plan to get there soon.

It may be theoretically possible to use git etc. over NFS but I wouldn't wish it to my worst enemy, and there is absolutely no reason to do it since nowadays even the cheapest laptop comes with half a terabyte or so of local storage.

I'm doing it over NFS4. It works fine, with some snags regarding pNFS/iSCSI.
And terabyte isn't really impressive nowadays. Disks are bigger, but so is the data.
And it helps to have one's data centralized.
Much less of unnecessary copies etc, which eventually have to be accunted for, even if you decide to delete them at some later point etc.

Shared filesystems are now used much more like a kind of FTP, where you download your stuff and then upload it again once you are done working on it - something that can (and should?) be replaced with webdav or sftp, rather than a pseudo-transparent network mount.

Coming trends tend to oppose that. CXL etc are to bring new levels of cheap connectivity for these and more advanced needs.

Besides NFS implementation in Linux has never been great and AFAIK Linux doesn't even support NFSv4 ACLs.

AFAIK they do. They also support SELinux context and IIRC node attributes, if the underlying fs supports them etc.

[numacross]

BTW, AFAIK most new 2.5G NICs do offer RDMA, which might be a good reason to get them, even if I run them at just 1Gb/s over old switch...

Do you have any examples? I have both Aquantia/Marvell aqn107 (10/5/2.5/1, but I doubt the 5/2.5 aqn108 is different in this regard) and Realtek 8125 (2.5) which do not mention RDMA anywhere. Intel's I225-V doesn't have it either it seems.

[Brane215]

Do you have any examples? I have both Aquantia/Marvell aqn107 (10/5/2.5/1, but I doubt the 5/2.5 aqn108 is different in this regard) and Realtek 8125 (2.5) which do not mention RDMA anywhere. Intel's I225-V doesn't have it either it seems.

Can't find it now. But I could swear I've seen it somewhere...