[gpfsug-discuss] Presentations from UG #6 - Feedback for IBM appreciated

Orlando Richards orlando.richards at ed.ac.uk
Tue Sep 25 14:41:57 BST 2012

On 25/09/12 14:05, Jez Tucker wrote:
> Hello all
>    Firstly can I thank all who attended UG #6.  We had a great turn out and the opportunity to network with more people from IBM was most welcome.
> I have uploaded the presentations from UG to this small, catchy URL: http://goo.gl/n1in1
> [Bar the SCCS presentation, awaiting clearance].
> Please have a read of the presentations.
> IBM Almaden Labs welcome your feedback regarding pNFS and Panache as well as FRQs etc.
> For instance, one FRQ idea banded around was a GRIO/QoS implementation for GPFS : E.G.: http://goo.gl/zjkN8
> It would be most helpful if a couple of lines use-case was alongside each of these.
> If these are messaged back to the list for healthy debate or sent to me directly I'll put them on the UG website for Almaden Labs to peruse/discuss with us.
> Regards,
>      Jez
> p.s. I'll also start to solicit previous presentations for UG < #5, so if you were a speaker, please get in touch.
> ---

Thanks for the great meeting Jez, and Claire et al at OCF.

On feature requests, I think one desirable feature request discussed at 
the meeting was for "better" performance monitoring tools.

A quick think through the things on my plate which would be eased with 
new/changed features in GPFS led me to this wishlist:

  - ability to change the designated NSD servers for an NSD without 
unmounting the filesystem everywhere

  - expansion of the AFM toolchain, including the following to assist 
with migration of data between filesystems:
    - ability to set a pre-existing fileset as a "cache" of an empty 
'home' fileset with AFM, allowing for a push of the data from the 
"cache" fileset/filesystem to the "home" target fileset/filesystem as a 
data migration strategy
    - ability to remove an AFM relationship between filesets, preserving 
data in the 'cache' fileset (and making it, independently, a 'live' fileset)
    - ability to "flip" the 'home'<->'cache' relationship, resulting in 
a flush from the new 'cache' fileset to the new 'home' fileset

  - better documentation (and, indeed, automation/automagic) on making 
best use of available memory within NSD servers

  - read caching of data blocks within an NSD server's memory (when 
acting in "server" mode in a multi-cluster environment where the client 
nodes do not have direct block access to the disks)

  - "hot file" identification tools/data for policy based HSM migration

  - some easy and non-invasive method for logging file and folder 
deletions (for the purposes of expiring backup data without using a 
separate database of files, in my case)

  - better licensing model (dare I say it - capacity based?)

I'd love to be able to change the blocksize on an existing filesystem 
too, but I imagine that's not possible.


    Dr Orlando Richards
   Information Services
IT Infrastructure Division
        Unix Section
     Tel: 0131 650 4994

The University of Edinburgh is a charitable body, registered in 
Scotland, with registration number SC005336.

More information about the gpfsug-discuss mailing list