[gpfsug-discuss] Metadata only system pool

Uwe Falke UWEFALKE at de.ibm.com
Tue Jan 23 20:35:56 GMT 2018


Hi, Kevin, 

(reproducing mainly what I've learned from Yuri)
a 512B inode can hold about 32 Disk Addresses (DAs) addressing 32 physical 
disk blocks (but mind: if max replication - not necessarily actual repl. 
-- is set to more than 1, that mans for example 16 logical blocks (R=2). A 
DA is 12 bytes. With a 4k inode you can expect the other 3.5kiB to contain 
also DAs throughout  (about 298)

So with R=2 (even if actual replication is r=1), a 4k inode can hold max 
about 330 DAs addressing 165 logical blocks 
The max file size w/out indirect addressing is thus 165 x blocksize (e.g. 
1320MiB @8MiB)

However, there could be other data structures in an inode (EAs) reducing 
the space available for DAs

Hence, YMMV

 
Mit freundlichen Grüßen / Kind regards

 
Dr. Uwe Falke
 
IT Specialist
High Performance Computing Services / Integrated Technology Services / 
Data Center Services
-------------------------------------------------------------------------------------------------------------------------------------------
IBM Deutschland
Rathausstr. 7
09111 Chemnitz
Phone: +49 371 6978 2165
Mobile: +49 175 575 2877
E-Mail: uwefalke at de.ibm.com
-------------------------------------------------------------------------------------------------------------------------------------------
IBM Deutschland Business & Technology Services GmbH / Geschäftsführung: 
Thomas Wolter, Sven Schooß
Sitz der Gesellschaft: Ehningen / Registergericht: Amtsgericht Stuttgart, 
HRB 17122 




From:   "Buterbaugh, Kevin L" <Kevin.Buterbaugh at Vanderbilt.Edu>
To:     gpfsug main discussion list <gpfsug-discuss at spectrumscale.org>
Date:   01/23/2018 08:26 PM
Subject:        Re: [gpfsug-discuss] Metadata only system pool
Sent by:        gpfsug-discuss-bounces at spectrumscale.org



Hi All, 

This is all making sense and I appreciate everyone?s responses ? and again 
I apologize for not thinking about the indirect blocks.

Marc - we specifically chose 4K inodes when we created this filesystem a 
little over a year ago so that small files could fit in the inode and 
therefore be stored on the metadata SSDs.

This is more of a curiosity question ? is it documented somewhere how a 4K 
inode is used?  I understand that for very small files up to 3.5K of that 
can be for data, but what about for large files?  I.e., how much of that 
4K is used for block addresses  (3.5K plus whatever portion was already 
allocated to block addresses??) ? or what I?m really asking is, given 4K 
inodes and a 1M block size how big does a file have to be before it will 
need to use indirect blocks?

Thanks again?

Kevin

On Jan 23, 2018, at 1:12 PM, Marc A Kaplan <makaplan at us.ibm.com> wrote:

If one were starting over, it might make sense to use a  smaller inode 
size.  I believe we still support 512, 1K, 2K.
Tradeoff with the fact that inodes can store data and EAs.




From:        "Uwe Falke" <UWEFALKE at de.ibm.com>
To:        gpfsug main discussion list <gpfsug-discuss at spectrumscale.org>
Date:        01/23/2018 04:04 PM
Subject:        Re: [gpfsug-discuss] Metadata only system pool
Sent by:        gpfsug-discuss-bounces at spectrumscale.org



rough calculation (assuming 4k inodes): 
350 x 10^6 x 4096 Bytes=1.434TB=1.304TiB. With replication that uses 
2.877TB or 2.308TiB 
As already mentioned here, directory and indirect blocks come on top. Even 

if you could get rid of a portion of the allocated and unused inodes that 
metadata pool appears bit small to me. 
If that is a large filesystem there should be some funding to extend it. 
If you have such a many-but-small-files system as discussed recently in 
this theatre, you might still beg for more MD storage but that makes than 
a larger portion of the total cost (assuming data storage is on HDD and md 

storage on SSD) and that again reduces your chances. 




Mit freundlichen Grüßen / Kind regards


Dr. Uwe Falke

IT Specialist
High Performance Computing Services / Integrated Technology Services / 
Data Center Services
-------------------------------------------------------------------------------------------------------------------------------------------
IBM Deutschland
Rathausstr. 7
09111 Chemnitz
Phone: +49 371 6978 2165
Mobile: +49 175 575 2877
E-Mail: uwefalke at de.ibm.com
-------------------------------------------------------------------------------------------------------------------------------------------
IBM Deutschland Business & Technology Services GmbH / Geschäftsführung: 
Thomas Wolter, Sven Schooß
Sitz der Gesellschaft: Ehningen / Registergericht: Amtsgericht Stuttgart, 
HRB 17122 




From:   "Buterbaugh, Kevin L" <Kevin.Buterbaugh at Vanderbilt.Edu>
To:     gpfsug main discussion list <gpfsug-discuss at spectrumscale.org>
Date:   01/23/2018 06:17 PM
Subject:        [gpfsug-discuss] Metadata only system pool
Sent by:        gpfsug-discuss-bounces at spectrumscale.org



Hi All, 

I was under the (possibly false) impression that if you have a filesystem 
where the system pool contains metadata only then the only thing that 
would cause the amount of free space in that pool to change is the 
creation of more inodes ? is that correct?  In other words, given that I 
have a filesystem with 130 million free (but allocated) inodes:

Inode Information
-----------------
Number of used inodes:       218635454
Number of free inodes:       131364674
Number of allocated inodes:  350000128
Maximum number of inodes:    350000128

I would not expect that a user creating a few hundred or thousands of 
files could cause a ?no space left on device? error (which I?ve got one 
user getting).  There?s plenty of free data space, BTW.

Now my system pool is almost ?full?:

(pool total)           2.878T                                   34M (  0%) 

      140.9M ( 0%)

But again, what - outside of me creating more inodes - would cause that to 

change??

Thanks?

Kevin

?
Kevin Buterbaugh - Senior System Administrator
Vanderbilt University - Advanced Computing Center for Research and 
Education
Kevin.Buterbaugh at vanderbilt.edu - (615)875-9633


_______________________________________________
gpfsug-discuss mailing list
gpfsug-discuss at spectrumscale.org
https://urldefense.proofpoint.com/v2/url?u=http-3A__gpfsug.org_mailman_listinfo_gpfsug-2Ddiscuss&d=DwIFAw&c=jf_iaSHvJObTbx-siA1ZOg&r=cvpnBBH0j41aQy0RPiG2xRL_M8mTc1izuQD3_PmtjZ8&m=8WgQlUnhzFycOZf-YvqYpdUkCiyEdRvWukE-KKRuFbE&s=aCywbK-1heVHPR8Fg74z9VxkGbNfCxMdtEKIDMWVIwI&e=





_______________________________________________
gpfsug-discuss mailing list
gpfsug-discuss at spectrumscale.org
https://urldefense.proofpoint.com/v2/url?u=http-3A__gpfsug.org_mailman_listinfo_gpfsug-2Ddiscuss&d=DwIFAw&c=jf_iaSHvJObTbx-siA1ZOg&r=cvpnBBH0j41aQy0RPiG2xRL_M8mTc1izuQD3_PmtjZ8&m=8WgQlUnhzFycOZf-YvqYpdUkCiyEdRvWukE-KKRuFbE&s=aCywbK-1heVHPR8Fg74z9VxkGbNfCxMdtEKIDMWVIwI&e=





_______________________________________________
gpfsug-discuss mailing list
gpfsug-discuss at spectrumscale.org
https://na01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fgpfsug.org%2Fmailman%2Flistinfo%2Fgpfsug-discuss&data=02%7C01%7CKevin.Buterbaugh%40vanderbilt.edu%7C77fefde14ec54e04b35708d5629550bd%7Cba5a7f39e3be4ab3b45067fa80faecad%7C0%7C0%7C636523315820548341&sdata=rbazh5e%2BxgHGvgF65VHTs9Hf4kk9EtUizsb19l5rr7U%3D&reserved=0
_______________________________________________
gpfsug-discuss mailing list
gpfsug-discuss at spectrumscale.org
http://gpfsug.org/mailman/listinfo/gpfsug-discuss






More information about the gpfsug-discuss mailing list