Now that the client side of pnfs is in the mainline kernel
I believe it is time to consider the inclusion of the server side.
I propose the following agenda for discussion
(order may change with no advance notice :)
* What's currently available in git://linux-nfs.org/~halevy/linux-pnfs.git
- What are the sub-projects
- How they relate to each other
* High-level design of the implementation
* Summary of generic changes in nfsd
* For each of the different sub-projects, briefly present:
- What does it do
- Benefits and Potential
- Limitations
- Status
- To-do
* Prerequisites for inclusion
* Discussion
Benny
On Mon, Jan 09, 2012 at 02:49:32PM +0200, Boaz Harrosh wrote:
> On 12/25/2011 06:30 AM, Benny Halevy wrote:
> > Now that the client side of pnfs is in the mainline kernel
> > I believe it is time to consider the inclusion of the server side.
We're also closing in on the basic 4.1 todo's, so I agree that it will
soon be time to talk about merging server-side pNFS.
> > I propose the following agenda for discussion
> > (order may change with no advance notice :)
> >
> > * What's currently available in git://linux-nfs.org/~halevy/linux-pnfs.git
> > - What are the sub-projects
> > - How they relate to each other
> >
> > * High-level design of the implementation
> >
> > * Summary of generic changes in nfsd
> >
> > * For each of the different sub-projects, briefly present:
> > - What does it do
> > - Benefits and Potential
> > - Limitations
> > - Status
> > - To-do
> >
> > * Prerequisites for inclusion
Understood that it may depend on where things stand in April, but: what
specifically do you think will be likely to require the attention of a
wider group of linux filesystem developers (ass opposed to just nfs
developers)?
--b.
> > * Discussion
> >
>
> Me too! This subject is close to my heart, as exofs is the most
> complete and advanced pNFSD base implementation. Last testing
> has demonstrated amazing performance and scalability. Saturating
> 10G from a single client and saturating 4*10G storage cluster
> from multiple clients.(Though there were problems with too many
> clients)
On Wed, Jan 11, 2012 at 03:28:53PM +0200, Boaz Harrosh wrote:
> On 01/11/2012 09:44 AM, Benny Halevy wrote:
> >>
> >> Understood that it may depend on where things stand in April, but: what
> >> specifically do you think will be likely to require the attention of a
> >> wider group of linux filesystem developers (ass opposed to just nfs
> >> developers)?
> >
> > I have nothing specific at this time. My thinking was to present the
> > what's and why's
> > of the prerequisites in high level and maybe provide an example.
> >
>
> At the time there where some reservations from people about some of the
> added hooks and vectors. If there are still, we would like to hear them
> and either address them or argue their validity.
OK.
> I think this talk should be expanded with Jeff's suggestions. here:
> > - RichACLs
> > - Share/Deny mode support on open
> > - mandatory locking that doesn't rely on weirdo file modes
...
> So the broader topic can be:
> "The things we want/need from the Kernel to make it exportable"
Agreed that no matter where we are there will likely be a long list of
thorny questions of that sort....
--b.
On Mon, Jan 9, 2012 at 9:25 PM, J. Bruce Fields <[email protected]> wrote:
> On Mon, Jan 09, 2012 at 02:49:32PM +0200, Boaz Harrosh wrote:
>> On 12/25/2011 06:30 AM, Benny Halevy wrote:
>> > Now that the client side of pnfs is in the mainline kernel
>> > I believe it is time to consider the inclusion of the server side.
>
> We're also closing in on the basic 4.1 todo's, so I agree that it will
> soon be time to talk about merging server-side pNFS.
>
>> > I propose the following agenda for discussion
>> > (order may change with no advance notice :)
>> >
>> > * What's currently available in git://linux-nfs.org/~halevy/linux-pnfs.git
>> > ? - What are the sub-projects
>> > ? - How they relate to each other
>> >
>> > * High-level design of the implementation
>> >
>> > * Summary of generic changes in nfsd
>> >
>> > * For each of the different sub-projects, briefly present:
>> > ? - What does it do
>> > ? - Benefits and Potential
>> > ? - Limitations
>> > ? - Status
>> > ? - To-do
>> >
>> > * Prerequisites for inclusion
>
> Understood that it may depend on where things stand in April, but: what
> specifically do you think will be likely to require the attention of a
> wider group of linux filesystem developers (ass opposed to just nfs
> developers)?
I have nothing specific at this time. My thinking was to present the
what's and why's
of the prerequisites in high level and maybe provide an example.
Benny
>
> --b.
>
>> > * Discussion
>> >
>>
>> Me too! This subject is close to my heart, as exofs is the most
>> complete and advanced pNFSD base implementation. Last testing
>> has demonstrated amazing performance and scalability. Saturating
>> 10G from a single client and saturating 4*10G storage cluster
>> from multiple clients.(Though there were problems with too many
>> clients)
On 01/11/2012 09:44 AM, Benny Halevy wrote:
>>
>> Understood that it may depend on where things stand in April, but: what
>> specifically do you think will be likely to require the attention of a
>> wider group of linux filesystem developers (ass opposed to just nfs
>> developers)?
>
> I have nothing specific at this time. My thinking was to present the
> what's and why's
> of the prerequisites in high level and maybe provide an example.
>
At the time there where some reservations from people about some of the
added hooks and vectors. If there are still, we would like to hear them
and either address them or argue their validity.
I think this talk should be expanded with Jeff's suggestions. here:
> - RichACLs
> - Share/Deny mode support on open
> - mandatory locking that doesn't rely on weirdo file modes
>
> It's always going to be hard for us to compete with dedicated
> appliances. Where Linux can shine though is in allowing for more
> innovative combinations.
>
> Being able to do active/active NFS serving from clustered filesystems,
> for instance is something that we can eventually attain but that would
> be harder to do in an appliance. This sort of discussion might also
> dovetail with Benny's proposal about pNFS serving.
So the broader topic can be:
"The things we want/need from the Kernel to make it exportable"
Thanks
Boaz
On 12/25/2011 06:30 AM, Benny Halevy wrote:
> Now that the client side of pnfs is in the mainline kernel
> I believe it is time to consider the inclusion of the server side.
>
> I propose the following agenda for discussion
> (order may change with no advance notice :)
>
> * What's currently available in git://linux-nfs.org/~halevy/linux-pnfs.git
> - What are the sub-projects
> - How they relate to each other
>
> * High-level design of the implementation
>
> * Summary of generic changes in nfsd
>
> * For each of the different sub-projects, briefly present:
> - What does it do
> - Benefits and Potential
> - Limitations
> - Status
> - To-do
>
> * Prerequisites for inclusion
>
> * Discussion
>
Me too! This subject is close to my heart, as exofs is the most
complete and advanced pNFSD base implementation. Last testing
has demonstrated amazing performance and scalability. Saturating
10G from a single client and saturating 4*10G storage cluster
from multiple clients.(Though there were problems with too many
clients)
> Benny
Thanks
Boaz