NEC's GFS for mainline kernels?

NEC's GFS for mainline kernels?

[Christoph Hellwig]

Hi Shibata-san,

is there a chance your team could release NEC's direct-to-SAN NFS code (*)
as a patch against stock kernels, too instead of only as part of the
kernel RPMs for NEC's IA64 servers?  The concept looks very nice, but I
think the integration with the scsi layer and the xfs/nfs code still
needs to be a lot cleaner which could benefit from some public review.

I was pretty suprised about GFS and some other interesting changes in
NEC's tree, it's a shame you hide such nice free software :)

	Christoph

(*) the name GFS clashes with another linux filesystem which is very
    confusing, btw..

Re: NEC's GFS for mainline kernels?

[Trond Myklebust]

>>>>> " " == Christoph Hellwig <hch@lst.de> writes:

     > Hi Shibata-san, is there a chance your team could release NEC's
     > direct-to-SAN NFS code (*) as a patch against stock kernels,
     > too instead of only as part of the kernel RPMs for NEC's IA64
     > servers?

Hi,

Is that some variant on the RDMA concept? (see http://www.rdmaconsortium.org)

If so, please note that there is an ongoing effort to standardize
RDMA-over-TCP via the IETF, and (of course) a parallel effort to
integrate it into a future minor revision of NFSv4 (see
http://www.snia.org/tech_activities/workgroups/nfs_rdma)

Although it's too late in the game to add generic support for RDMA
into 2.5.x/2.6.x, this type of project is certainly a very interesting
candidate for inclusion in future kernels.

Cheers,
  Trond

Re: NEC's GFS for mainline kernels?

[David Chow]

Trond Myklebust wrote:

>>>>>>" " == Christoph Hellwig <hch@lst.de> writes:
>>>>>>            
>>>>>>
>
>     > Hi Shibata-san, is there a chance your team could release NEC's
>     > direct-to-SAN NFS code (*) as a patch against stock kernels,
>     > too instead of only as part of the kernel RPMs for NEC's IA64
>     > servers?
>
>Hi,
>
>Is that some variant on the RDMA concept? (see http://www.rdmaconsortium.org)
>
>If so, please note that there is an ongoing effort to standardize
>RDMA-over-TCP via the IETF, and (of course) a parallel effort to
>integrate it into a future minor revision of NFSv4 (see
>http://www.snia.org/tech_activities/workgroups/nfs_rdma)
>
>Although it's too late in the game to add generic support for RDMA
>into 2.5.x/2.6.x, this type of project is certainly a very interesting
>candidate for inclusion in future kernels.
>
>Cheers,
>  Trond
>
>  
>
I am also confused. Are you talking about the DAFS project? Which file 
system are you talking about?

regards,
David

Re: NEC's GFS for mainline kernels?

[Trond Myklebust]

>>>>> " " == David Chow <davidchow@shaolinmicro.com> writes:

     > I am also confused. Are you talking about the DAFS project? 

Me? No.

     > Which file system are you talking about?

My posting related to efforts to add RDMA support to the NFS protocol
since this is what I gathered the NEC GFS effort is all about. See the
links I posted.

DAFS is interesting too, and does indeed have built in support for
RDMA, but the DAFS protocol is not the same as NFS.

Cheers,
  Trond

Re: NEC's GFS for mainline kernels?

[Christoph Hellwig]

On Tue, Jan 28, 2003 at 05:34:17PM +0100, Trond Myklebust wrote:
> >>>>> " " == Christoph Hellwig <hch@lst.de> writes:
> 
>      > Hi Shibata-san, is there a chance your team could release NEC's
>      > direct-to-SAN NFS code (*) as a patch against stock kernels,
>      > too instead of only as part of the kernel RPMs for NEC's IA64
>      > servers?
> 
> Hi,
> 
> Is that some variant on the RDMA concept? (see http://www.rdmaconsortium.org)
> 
> If so, please note that there is an ongoing effort to standardize
> RDMA-over-TCP via the IETF, and (of course) a parallel effort to
> integrate it into a future minor revision of NFSv4 (see
> http://www.snia.org/tech_activities/workgroups/nfs_rdma)

No, it doesn't go over the network but directly to fibrechannel.
See the diff (2.4.18 plain -> 2.4.18 nec) that I've put on nl.linux.org:

	http://nl.linux.org/~hch/patch-2.4.18-nec3.2.bz2

Re: NEC's GFS for mainline kernels?

[Takashi Yanagawa]

Thank you for having interest in GFS. 

I who takes charge of GFS function reply in place of Shibata-san. 

Though it is completed functionally and it works, GFS doesn't reach the 
level that it can be introduced to the public, because GFS codes have to
be cleaner as you mentioned below. 


On Tue, 28 Jan 2003 16:32:41 +0100
Christoph Hellwig <hch@lst.de> wrote:

> Hi Shibata-san,
> 
> is there a chance your team could release NEC's direct-to-SAN NFS code (*)
> as a patch against stock kernels, too instead of only as part of the
> kernel RPMs for NEC's IA64 servers?  The concept looks very nice, but I
> think the integration with the scsi layer and the xfs/nfs code still
> needs to be a lot cleaner which could benefit from some public review.
> 
> I was pretty suprised about GFS and some other interesting changes in
> NEC's tree, it's a shame you hide such nice free software :)
> 
> 	Christoph
> 
> (*) the name GFS clashes with another linux filesystem which is very
>     confusing, btw..
> 

-------------------------------------------------------------------
    Takashi Yanagawa           e-mail: t-yanagawa@bc.jp.nec.com