For future filesystem compatibility, we are wondering if there are any Lustre MDT filesystems in existence that have 2B or more total inodes? This is fairly unlikely, because it would require an MDT filesystem that is > 8TB in size (which isn''t even supported yet) and/or has been formatted with specific options to increase the total number of inodes. This can be checked with "dumpe2fs -h /dev/{mdtdev} | grep ''Inode count''" on the MDT. If there are issues with privacy, please email me directly. Otherwise, I think people would be interested to know what the larger MDT sizes are (even if they aren''t as high as 2B inodes). Cheers, Andreas -- Andreas Dilger Sr. Staff Engineer, Lustre Group Sun Microsystems of Canada, Inc.
We have ~1.4 billion inodes and we''re using about 50 million. We use lots of small files so I took our highest utilization ever and doubled it. Better safe than sorry. We did format with an inode every 1024 bytes because we use no striping. -- Daniel Leaberry Senior Systems Administrator iArchives Inc. Andreas Dilger wrote:> For future filesystem compatibility, we are wondering if there are any > Lustre MDT filesystems in existence that have 2B or more total inodes? > > This is fairly unlikely, because it would require an MDT filesystem > that is > 8TB in size (which isn''t even supported yet) and/or has been > formatted with specific options to increase the total number of inodes. > > This can be checked with "dumpe2fs -h /dev/{mdtdev} | grep ''Inode count''" > on the MDT. > > > If there are issues with privacy, please email me directly. Otherwise, > I think people would be interested to know what the larger MDT sizes > are (even if they aren''t as high as 2B inodes). > > Cheers, Andreas > -- > Andreas Dilger > Sr. Staff Engineer, Lustre Group > Sun Microsystems of Canada, Inc. > > _______________________________________________ > Lustre-discuss mailing list > Lustre-discuss at lists.lustre.org > http://lists.lustre.org/mailman/listinfo/lustre-discuss
We have ~1.4 billion inodes and we''re using about 50 million. We use lots of small files so I took our highest utilization ever and doubled it. Better safe than sorry. We did format with an inode every 1024 bytes because we use no striping. -- Daniel Leaberry Senior Systems Administrator iArchives Inc. Andreas Dilger wrote:> For future filesystem compatibility, we are wondering if there are any > Lustre MDT filesystems in existence that have 2B or more total inodes? > > This is fairly unlikely, because it would require an MDT filesystem > that is > 8TB in size (which isn''t even supported yet) and/or has been > formatted with specific options to increase the total number of inodes. > > This can be checked with "dumpe2fs -h /dev/{mdtdev} | grep ''Inode count''" > on the MDT. > > > If there are issues with privacy, please email me directly. Otherwise, > I think people would be interested to know what the larger MDT sizes > are (even if they aren''t as high as 2B inodes). > > Cheers, Andreas > -- > Andreas Dilger > Sr. Staff Engineer, Lustre Group > Sun Microsystems of Canada, Inc. > > _______________________________________________ > Lustre-discuss mailing list > Lustre-discuss at lists.lustre.org > http://lists.lustre.org/mailman/listinfo/lustre-discuss
> Otherwise, > I think people would be interested to know what the larger MDT sizes > are (even if they aren''t as high as 2B inodes). >I have three clusters of 400M inodes each. MDT formated with an inode size of 256 and one inode per 1024 byte: ''-I 256 -i 1024''. The cluster is split in three primarily to keep an eventual restore time low but ideally this would be one big cluster with 1.2G inodes. To other users: Note the inode size is less than recommended. My stripecount is one so this is feasible without a performance hit but Andreas warned me about future features which might add more user attributes. /Jakob
D. Marc Stearman
2008-Jun-19 15:13 UTC
[Lustre-discuss] [Lustre-devel] maximum MDT inode count
On our filesystem that was built to support BGL, we have 2B inodes with 2T of capacity. Inode count: 2097315840 Inodes per group: 32768 Inode blocks per group: 4096 Inode size: 512 We explicitly chose this number and put it in the XML when the filesystem was built. Then the users changed their I/O pattern and did not end up creating nearly as many files as we estimated. Since then we have created file systems with a smaller number of inodes to reduce fsck time. (Not really an issue as much now, though with the fast fsck feature). -Marc ---- D. Marc Stearman LC Lustre Administration Lead marc at llnl.gov 925.423.9670 Pager: 1.888.203.0641 On Jun 17, 2008, at 10:10 AM, Andreas Dilger wrote:> For future filesystem compatibility, we are wondering if there are any > Lustre MDT filesystems in existence that have 2B or more total inodes? > > This is fairly unlikely, because it would require an MDT filesystem > that is > 8TB in size (which isn''t even supported yet) and/or has been > formatted with specific options to increase the total number of > inodes. > > This can be checked with "dumpe2fs -h /dev/{mdtdev} | grep ''Inode > count''" > on the MDT. > > > If there are issues with privacy, please email me directly. > Otherwise, > I think people would be interested to know what the larger MDT sizes > are (even if they aren''t as high as 2B inodes). > > Cheers, Andreas > -- > Andreas Dilger > Sr. Staff Engineer, Lustre Group > Sun Microsystems of Canada, Inc. > > _______________________________________________ > Lustre-devel mailing list > Lustre-devel at lists.lustre.org > http://lists.lustre.org/mailman/listinfo/lustre-devel