Subject: Re: newfs defaults
To: John Klos <email@example.com>
From: David Laight <firstname.lastname@example.org>
Date: 10/14/2005 20:48:45
On Wed, Oct 12, 2005 at 07:26:27PM -0700, John Klos wrote:
> Playing with a fibre channel RAID array, I noticed that newfs on a 2 TB
> slice defaults to block size of 16384 meaning that it would create 11,374
> cylinder groups which seems a bit excessive. Is this normal and desirable
> for large disks?
There is little choice....
The allocation bitmaps have to fit into a single filesystem block.
(something else that FFSv2 should have changed...)
So a 'cylinder group' can only describe 'block_sz*8' fragments (ignoring
red tape and inodes) or 'block_sz*8*frag_sz' bytes.
With 16k blocks and 2k fragments this is ~256MB requiring the 11k cylinder
Using 32k blocks would reduce the number of cylinder groups to about 3000.
Clearly you could use 64k blocks and 64k fragments, giving 'only' 750
cylinder groups - but that wouldn't be bootable.
I have considered changing newfs to output a '.' for each cylinder group
after the first line of superblock sector numbers.
David Laight: email@example.com