[LSF/MM TOPIC] Network and Cluster File System Pain Points

[LSF/MM TOPIC] Network and Cluster File System Pain Points

[Steve French]

There are various topics particularly important to network and cluster
file systems:
     - additional metadata on files, directories, mount points (statx,
enhancements to the mount API, ACLs etc.)
     - performance related features (leases/delegation, RDMA
integration, reflink/clone extensions, handling sideeffects of long
latency network connections on common fs operations by better
aggregating series of related operations together)
     - security features
     - mapping features available in various network and cluster FS
protocols to Linux

 (in the past we also have discussed richacl, and a few features that
would help Samba server).

-- 
Thanks,

Steve

Re: [LSF/MM TOPIC] Network and Cluster File System Pain Points

[Chuck Lever]

> On Feb 8, 2019, at 10:59 AM, Steve French <smfrench@gmail.com> wrote:
> 
> There are various topics particularly important to network and cluster
> file systems:
>     - additional metadata on files, directories, mount points (statx,
> enhancements to the mount API, ACLs etc.)
>     - performance related features (leases/delegation, RDMA
> integration, reflink/clone extensions, handling sideeffects of long
> latency network connections on common fs operations by better
> aggregating series of related operations together)
>     - security features
>     - mapping features available in various network and cluster FS
> protocols to Linux
> 
> (in the past we also have discussed richacl, and a few features that
> would help Samba server).

I would prefer to focus on just one or two specific topics rather
than walking down a laundry list of items that we've visited in
previous years. Are there one or two particular new issues where
you need some group decision-making?

Eg, the file utilities topic seems like a specific topic, interesting
for interoperation with cloud provider file systems that are typically
large distributed block stores, and thus are slow at doing metadata
updates but capable of great aggregate update bandwidth.


--
Chuck Lever
chucklever@gmail.com

Re: [LSF/MM TOPIC] Network and Cluster File System Pain Points

[Steve French]

On Fri, Feb 8, 2019 at 12:57 PM Chuck Lever <chucklever@gmail.com> wrote:
>
>
>
> > On Feb 8, 2019, at 10:59 AM, Steve French <smfrench@gmail.com> wrote:
> >
> > There are various topics particularly important to network and cluster
> > file systems:
> >     - additional metadata on files, directories, mount points (statx,
> > enhancements to the mount API, ACLs etc.)
> >     - performance related features (leases/delegation, RDMA
> > integration, reflink/clone extensions, handling sideeffects of long
> > latency network connections on common fs operations by better
> > aggregating series of related operations together)
> >     - security features
> >     - mapping features available in various network and cluster FS
> > protocols to Linux
> >
> > (in the past we also have discussed richacl, and a few features that
> > would help Samba server).
>
> I would prefer to focus on just one or two specific topics rather
> than walking down a laundry list of items that we've visited in
> previous years. Are there one or two particular new issues where
> you need some group decision-making?
>
> Eg, the file utilities topic seems like a specific topic, interesting
> for interoperation with cloud provider file systems that are typically
> large distributed block stores, and thus are slow at doing metadata
> updates but capable of great aggregate update bandwidth.

I agree - I had submitted a separate one for the copy problem (where I see
four or five obvious enhancements that could use group discussion
and from looking at github projects see some earlier presumably
abandoned updates to cp to address some of them).

But for additional file utilities - it would make sense to have a short session
to discuss those - NFS would likely have similar requirements to SMB3
(and even some other fs) for that.  What other file utilities besides the
copy ones are the ones that NFS users complain most about not having or
needing to improve?
-- 
Thanks,

Steve