|
After the last episode
with Solaris and ZFS, I now attempted to emulate what would have been
done with ZFS, I created a meta device to stripe across the two RAID5s of
the xraid (via metadb(1M) and metainit(1M)).
This yielded a 5.4 TB disk, which can lateron be extended via
metainit(1M) and growfs(1M).
I then proceeded to create and populate a regular UFS filesystem on the
resulting disk, only to remember lateron that per default
newfs(1M) enforces a ridicolous inode density on filesystems
larger than one 1TB. (The 5.4 TB UFS filesystem created on Solaris
provides for 5976768 inodes; my IRIX XFS filesystems that I intend
to migrate contain 385056768 inodes (367GB total space) plus 2 x
1085618176 inodes (2 x 1TB total space). If I want to place all this on
the single striped volume, I'll need 2556293120 inodes to just get as
many as I already have; on a 5.4 TB filesystem (ie more than double the
size), I'll likely need more inodes, too.)
I had to search high and lo through the interwebbing until I found this
thread about the same problem. It appears that indeed the only way to
create a filesystem of this size with an acceptable inode density is to
compile your own 'mkfs' from OpenSolaris (for
example: from
here) and then run ./mkfs -F ufs /dev/md/rdsk/d1
11721015296 128 48 8192 8192 -1 1 1 16384 t 0 -1 ./mkfs -F ufs
/dev/md/rdsk/d1 11721015296 128 48 8192 8192 -1 1 1 8192 t 0
-1 I don't even want to think about how long this takes to fsck, if
it ever has to! So it looks like I can't have just one giant
filesystem and instead will need to create many smaller filesystems. How
old-fashioned! ;-/
July 26, 2006
|