Re: #blocks per group too big: 37265

[Theodore Tso <tytso@MIT.EDU>]

On Thu, Apr 03, 2008 at 06:19:04PM +0200, Paul Slootman wrote:
> Hi,
> I wanted to try out ext4 on my shiny new 9+TB RAID5 device
> (11 x 1TB disks in md raid5).
> 
> I obtained the 1.39-tyt3 version of e2fsprogs, and did:
> 
> ./mkfs.ext3 -j -m 0 -N 1000000000 -O dir_index,filetype,resize_inode -E stride=65536,resize=5120000000 -J device=/dev/mapper/vg11-md15--journal -L data2 /dev/md15
> 
> (If using a separate device for the journal is inadvisable, please let
> me know; this is on a different set of spindles that md15 is running on.)
> 
> The stride was calculated from the 64k chunk of the raid5 device.
> Mainly a guess, as I couldn't find any clear reference on how to plug in
> the values to fill this in.

The stride parameter is the problem.  Newer versions of e2fsprogs
don't allow a stride parameter which is too big.  If you want to do
the perfect calculation, you take the 64k chunk size, and divide it by
the 4k blocksize to yield a stride parameter of 16.  Actually, though,
simply using a non-zero stride size is actually good enough --- and if
you have a even number of RAID-5 disks, you might not need this
parameter at all.  (It's only purpose is to perturb the location of
the block bitmaps so that all of the bitmaps don't end up on a single
hard drive.)

BTW, we will be making a new snapshot for people who want to test ext4
soon....

					- Ted