[gpfsug-discuss] File placement policy based on creation and modification times
Ott Oopkaup
ott.oopkaup at ut.ee
Fri Jun 30 13:18:28 BST 2023
Hi
while maybe not the solution to the exact problem, GPFS does allow heat
based tiering which seems to me like a more correct way to ensure
efficient utilisation of fast SSD space.
https://www.ibm.com/docs/en/storage-scale/5.0.5?topic=scale-file-heat-tracking-file-access-temperature
Best,
Ott Oopkaup
University of Tartu, High Performance Computing Centre
Systems Administrator
On 6/29/23 12:01, Alec wrote:
> Yeah that kind of placement isn't possible, because you can only use
> attributes you know at the time of inode creation. When a file is
> created it's created with the current timestamp and then updated
> (usually after the copy finishes). If the majority of your data is
> going to be older than 365 days you may want to make your file
> placement default to your pool2, and then when you've finished copying
> all your older data, and want to freshen your data, update the
> placement policy it to the proper pool so new data hits the high speed
> disk.
>
> You can use a file path/name in the placement policy and some copy
> engines do give inodes temporary names before giving them their proper
> name... like rsync will start a file off with a . (and add a random
> suffix) until the file is completely transferred, then move it to the
> destination filename, then it will update the date, time, and
> ownership on that inode. So you could have your placement engine put
> anything starting with a . into your pool2 and then migrate fresher
> files back up to your higher tiered storage if desired.
>
> Not sure if any of that helps.
>
> Alec
>
> On Wed, Jun 28, 2023 at 10:12 PM Timm Stamer
> <timm.stamer at uni-oldenburg.de> wrote:
>
> Hello Prasad,
>
> we use this in our weekly policy run:
>
> RULE 'migrate cold data' MIGRATE FROM POOL 'system' TO POOL 'data'
> WHERE CURRENT_TIMESTAMP - ACCESS_TIME > INTERVAL '30' DAYS
>
>
> I do not know if a direct placement based on timestamps is possible.
>
>
>
> Kind regards
> Timm Stamer
>
>
> Am Mittwoch, dem 28.06.2023 um 23:18 +0000 schrieb Prasad Surampudi:
> >
> >
> >
> > ACHTUNG! Diese E-Mail kommt von Extern!WARNING! This email
> originated
> > off-campus.
> >
> >
> >
> >
> > Can we setup a file placement policy based on creating and
> > modification times when copying data from Windows into GPFS? It
> looks
> > like the placement policy only accepts only CREATION_TIME and not
> > MODIFICATION_TIME or ACCESS_TIME. If I try to use these, I get
> > message saying these are not supported in the context (placement?)
> > But even the policy with CREATION_TIME is not working properly. We
> > wanted files with CREATION_TIME which is 365 days ago go to a
> > different pool other than ‘system’. But when we copy files it is
> > dumping all files into system pool. But the creation time is looks
> > correct on the file after copied into GPFS. Does it check file
> > CREATION_TIME when a file gets copied over to GPFS?
> >
> > Here is the placement policy:
> > RULE ‘tiering’ SET POOL ‘pool2’
> > WHERE ( DAYS(CURRENT_TIMESTAMP) - DAYS(CREATION_TIME) > 365 )
> > RULE ‘default’ SET POOL ‘system’
> >
> >
> >
> > Logo
> >
> > Description automatically generated
> >
> > Prasad Surampudi | Sr. Systems Engineer
> > prasad.surampudi at theatsgroup.com | 302.419.5833
> >
> > Innovative IT consulting & modern infrastructure solutions
> > www.theatsgroup.com <http://www.theatsgroup.com>
> > _______________________________________________
> > gpfsug-discuss mailing list
> > gpfsug-discuss at gpfsug.org <http://gpfsug.org>
> > http://gpfsug.org/mailman/listinfo/gpfsug-discuss_gpfsug.org
> _______________________________________________
> gpfsug-discuss mailing list
> gpfsug-discuss at gpfsug.org <http://gpfsug.org>
> http://gpfsug.org/mailman/listinfo/gpfsug-discuss_gpfsug.org
>
>
> _______________________________________________
> gpfsug-discuss mailing list
> gpfsug-discuss at gpfsug.org
> http://gpfsug.org/mailman/listinfo/gpfsug-discuss_gpfsug.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://gpfsug.org/pipermail/gpfsug-discuss_gpfsug.org/attachments/20230630/b03b9496/attachment.htm>
More information about the gpfsug-discuss
mailing list