[gpfsug-discuss] Flash for metadata

[Yuri L Volobuev]

GPFS will issue 4K IOs if the file system is 4K-aligned, which is the
default for 4.1.1+, and may issue requests as small as 512 bytes otherwise.
If a given array presents 4K physical block size to the world, you won't be
able to format a non-4K-aligned file system which puts metadata on that
disk.  It is generally believed that 4K IO size produces optimal
performance with flash-based storage in general, because this is the native
physical block size, and 512-byte sector emulation entails doing a
read-modify-write.  We don't have the data on how much difference the
sector size actually makes.  If in doubt, I would recommend going with 4K
sectors and a 4K-aligned file system.

yuri



From:	"Simon Thompson (Research Computing - IT Services)"
            <S.J.Thompson at bham.ac.uk>
To:	"'gpfsug-discuss at spectrumscale.org'"
            <gpfsug-discuss at spectrumscale.org>,
Date:	05/16/2016 02:48 AM
Subject:	Re: [gpfsug-discuss] Flash for metadata
Sent by:	gpfsug-discuss-bounces at spectrumscale.org




Hi Paul,

Thanks, interestingly the FS-900 only does raid5. I get no choice about
that, I do get a sector size choice when I create a lun on it...

I was also wondering how gpfs would do a metadata update. Would it be smart
enough to know the sector size was 512b and only modify that, or would it
always do a read modify write on the whole 4k inode?

Simon

-----Original Message-----
From: Sanchez, Paul [Paul.Sanchez at deshaw.com]
Sent: Monday, May 16, 2016 10:28 AM GMT Standard Time
To: gpfsug-discuss at spectrumscale.org
Subject: Re: [gpfsug-discuss] Flash for metadata

Simon,

I can’t speak specifically to the FS-900, since I generally use Dell
Compellent and NetApp EF arrays for flash today.  But I also take care to
ensure that the system pool metadata NSDs use RAID10 (or equivalent) rather
than RAID5/6.  For small 4K metadata writes, on a 4K blocksize LUN this
could result in optimal writes.  For larger native blocksize LUNs, it
eliminates the parity calculation for read-modify-write operations.

Thanks,
Paul

From: gpfsug-discuss-bounces at spectrumscale.org [
mailto:gpfsug-discuss-bounces at spectrumscale.org] On Behalf Of Simon
Thompson (Research Computing - IT Services)
Sent: Monday, May 16, 2016 9:44 AM
To: gpfsug-discuss at spectrumscale.org
Subject: [gpfsug-discuss] Flash for metadata

Hi all

I'm currently about to deploy some flash systems (FS-900) to take our
metadata for one of our Spectrum Scale file-systems. We use 4k inodes for
this file-system.

On the FS-900, when creating a volume, we get a choice of 512b or 4k
sectors. Does anyone have a suggestion on this? On a spinning disk RAID
array, I get what a strip size is and we'd use multiples of the stripe
where possible for the block size.

Is the sector size on the FS-900 the smallest chunk that can be written
out? And so would it make sense to use 4k sectors for 4k GPFS inodes right?

Related. In the deployment guide for the FS-900 arrays, if one is using a
file-system (e.g. XFS) on the array the guide talks about aligning the
partitions with with array, when GPFS is used on a storage array, does it
automatically work out the sector alignment?


Thanks

Simon_______________________________________________
gpfsug-discuss mailing list
gpfsug-discuss at spectrumscale.org
http://gpfsug.org/mailman/listinfo/gpfsug-discuss

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://gpfsug.org/pipermail/gpfsug-discuss_gpfsug.org/attachments/20160524/3a332e95/attachment-0002.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: graycol.gif
Type: image/gif
Size: 105 bytes
Desc: not available
URL: <http://gpfsug.org/pipermail/gpfsug-discuss_gpfsug.org/attachments/20160524/3a332e95/attachment-0002.gif>