Re: [nfsv4] New draft for working group charter

[David Noveck <[e-mail]>]

> How about:
>
>  "features that facilitate administration of NFS storage in
>  scale-out virtualization environments"

OK

> That seems like what all the pNFS layout types do.

Except for pNFS file.

> "using already existing protocols" is not terribly helpful.
> That's also what other pNFS layout types do.

I think the import is intended to be that no control
protocol is required.

> Perhaps one of the Primary Data engineers can give us a
> crisp one-sentence mission statement for flexfiles.

I think they will, although we might have to wait until IETF-99.

> If not, then you could make a generic statement about
> broadening the repertoire of layout types to include
> new storage protocols such as NVMe on Fabrics, and
> leave it at that. I'd hate to do that, though, since
> a lot of work has gone into flexfiles.

A lot of work has gone into it and it is one of the few
working group documents that have not been through
WGLC.  Therefore, it has to be represented in the updated
charter, if possible. But I think I need Primary Data's input to
put something about it in the charter draft.

I think Christoph's NVMe work would be adequately represented
by your statement about use of new storage technologies.
You can judge how that works once I send raft #3 out.
We can add something flex-files-related once Primary
Data can suggest something appropriate to add.






On Sun, May 14, 2017 at 10:57 AM, Chuck Lever <[e-mail]>
wrote:

>
> > On May 13, 2017, at 4:08 PM, David Noveck <[e-mail]> wrote:
> >
> > > > > enabling use of NFS in warehouse virtualization environments,
> > > > >
> > > > I don't object to this but I'm not clear what work is beIng referred
> to.
> > > > Please clarify.
> >
> > > NFSv4.2 features such as ALLOCATE and SEEK_HOLE and space
> > > reservations and S2SC. These are very useful when preparing
> > > or moving virtual disk images, or when thin provisioning is
> > > in play.
> >
> > I get it now.  I think I was distracted by the word "warehouse".
> >
> > The IESG could be as well.  How about:
> >
> > Features to make NFS more suitable for use in virtualized
> > environments.
>
> That could mean running on my desktop. I'm specifically talking
> about features that are designed to make administering a large
> number of guests and hosts more straightforward.
>
> How about:
>
>  "features that facilitate administration of NFS storage in
>   scale-out virtualization environments"
>
>
> > > The re-treading of NFS/RDMA. NFS/TCP is typically slow in
> > > guests, but NFS/RDMA works in guests as well as NFS/TCP
> > > works on bare metal.
> >
> > Cool.  it seems to me that there is room for improvement in that
> > result.  That would primarily be implementation work, but the
> > protocol might be able to help as well.
> >
> > > > > and extending
> > > > > the life of legacy storage by enabling pNFS access to it?
> > >
> > >>  I"m not sure what this refering to either.
> >
> > > The pNFS flexfiles layout type. Feel free to adjust this
> > > item if you feel my simple summary misses the mark.
> >
> > It is not that you missed the mark.  It is just when i saw
> > "legacy storage", I thought "spinning metal".
> >
> > I'm thinking that a better way to say this might be:
> >
> > Provide NFSv4 to files stored using other file access
> > protocol, through the definiton of appropriate pNFS mapping
> > types.
>
> That seems like what all the pNFS layout types do.
>
> The Abstract of the flexfiles I-D says:
>
>    The flexible file layout type is defined in this
>    document as an extension to pNFS which allows the use of storage
>    devices in a fashion such that they require only a quite limited
>    degree of interaction with the metadata server, using already
>    existing protocols.  Client side mirroring is also added to provide
>    replication of files.
>
> "using already existing protocols" is not terribly helpful.
> That's also what other pNFS layout types do.
>
> Perhaps one of the Primary Data engineers can give us a
> crisp one-sentence mission statement for flexfiles.
>
> If not, then you could make a generic statement about
> broadening the repertoire of layout types to include
> new storage protocols such as NVMe on Fabrics, and
> leave it at that. I'd hate to do that, though, since
> a lot of work has gone into flexfiles.
>
> --
> Chuck Lever
>
>
>
>