I recently asked Which Filesystem should I use for my new 10TB partition. Thanks for all the comments and I want to answer a few of them.
There were a few recommendations (1, 2) to use XFS. I hear it a lot, but unfortunately it has never been as stable as EXT3 when I tried to use it in the past. It is important to remember that I am running it on a mirror server and after all the years maintaining our mirror server I have learned that the load which is generated by around 1500 users accessing different files at the same time is really different to anything else. I need to have a filesystem which is always stable and in my experience this is only fulfilled by EXT3. Because if something breaks I have so much data, that I cannot make a backup and restore it.
The other reason to not use XFS is that I am running a CentOS kernel and XFS support is only available through kmod-xfs and I have no idea how well maintained it is and if somebody is backporting all the important bugfixes to 2.6.18 (the CentOS kernel version).
Somebody said I should use ZFS and mount it via NFS. I was so happy a few years ago to get rid of NFS in our mirror server setup and I really do not want to have it back. I do not know if other mirrors are using NFS, but it never was any good in our setup. Currently we are using iSCSI if we have to mount something from a remote host. It works pretty good but does not help here because I would still have to create the filesystem locally.
I was happy to hear that Fedora infrastructure uses 10TB partitions with EXT3 (3, 4) and that was exactly the information I wanted to hear.
I have now created a 10TB partition using mke2fs -F -j /dev/sdf1 and it only took about 18 hours
. The RAID is still initializing in the background which makes all accesses a bit slower.
It is now mounted any I am starting to sync the data from the temporary iSCSI volume back to the new RAID:
/dev/sdf1 9.0T 26G 9.0T 1% /ftp/pub/.3
I have also benchmarked it with bonnie++ (bonnie++ -d bonnie -s 80g -m rhlx01 -u 123 -g 321), once with the background initializing still running:
Version 1.03 ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
rhlx01 80G 40390 92 62845 30 17201 4 27259 43 31459 3 98.5 0
and once in a completely initialized state:
Version 1.03 ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
rhlx01 80G 44377 96 95589 40 36061 7 32464 51 70583 7 242.7 0
Those values are pretty much the same as with the old RAID.