[gpfsug-discuss] Multiple Block Sizes in a Filesystem (Was: Online data migration tool)

Bryan Banister bbanister at jumptrading.com
Mon Jan 8 23:48:57 GMT 2018

Hey Aaron... I have been talking about the same idea here and would say it would be a massive feature and management improvement.

I would like to have many GPFS storage pools in my file system, each with tuned blocksize and subblock sizes to suite the application, using independent filesets and the data placement policy to store the data in the right GPFS storage pool.  Migrating the data with the policy engine between these pools as you described would be a lot faster and a lot safer than trying to migrate files individually (like with rsync).

NSDs can only belong to one storage pool, so I don't see why the block allocation map would be difficult to manage in this case.


-----Original Message-----
From: gpfsug-discuss-bounces at spectrumscale.org [mailto:gpfsug-discuss-bounces at spectrumscale.org] On Behalf Of Aaron Knister
Sent: Monday, January 08, 2018 4:57 PM
To: gpfsug main discussion list <gpfsug-discuss at spectrumscale.org>
Subject: [gpfsug-discuss] Multiple Block Sizes in a Filesystem (Was: Online data migration tool)

Note: External Email

I was thinking some more about the >32 subblock feature in scale 5.0. As
mentioned by IBM the biggest advantage of that feature is on filesystems
with large blocks (e.g. multiple MB). The majority of our filesystems
have a block size of 1MB which got me thinking... wouldn't it be nice if
they had a larger block size (there seem to be compelling performance
reasons for large file I/O to do this).

I'm wondering, what's the feasibility is of supporting filesystem pools
of varying block sizes within a single filesystem? I thought allocation
maps exist on a per-pool basis which gives me some hope it's not too hard.

If one could do this then, yes, you'd still need new hardware to migrate
to a larger block size (and >32 subblocks), but it could be done as part
of a refresh cycle *and* (and this is the part most important to me) it
could be driven entirely by the policy engine which means storage admins
are largely hands off and the migration is by and large transparent to
the end user.

This doesn't really address the need for a tool to address a filesystem
migration to 4k metadata blocks (although I wonder if something could be
done to create a system_4k pool that contains 4k-aligned metadata NSDs
where key data structures get re-written during a restripe in a
4k-aligned manner, but that's really grasping at straws for me).


Aaron Knister
NASA Center for Climate Simulation (Code 606.2)
Goddard Space Flight Center
(301) 286-2776
gpfsug-discuss mailing list
gpfsug-discuss at spectrumscale.org


Note: This email is for the confidential use of the named addressee(s) only and may contain proprietary, confidential or privileged information. If you are not the intended recipient, you are hereby notified that any review, dissemination or copying of this email is strictly prohibited, and to please notify the sender immediately and destroy this email and any attachments. Email transmission cannot be guaranteed to be secure or error-free. The Company, therefore, does not make any guarantees as to the completeness or accuracy of this email or any attachments. This email is for informational purposes only and does not constitute a recommendation, offer, request or solicitation of any kind to buy, sell, subscribe, redeem or perform any type of transaction of a financial product.

More information about the gpfsug-discuss mailing list