Re: [RFC 0/13] extents and 48bit ext3

Theodore Tso wrote:
> On Fri, Jun 09, 2006 at 11:11:31PM -0400, Jeff Garzik wrote:
>> It's an example of ext2 being bandaided to do something it was never 
>> originally designed to do.  If online resizing had been planned from the 
>> start, allocating new inode tables on the fly would be trivial, as it is 
>> in JFS/NTFS/...
> 
> And once again this has *nothing* to do with inode allocation, or
> dynamic allocation of inode tables.  Your "performance issue" has to
> do with a difference in blocksizes.  If you ext2/3 to pass your silly
> test, then upgrade to the latest e2fsprogs and install the following
> /etc/mke2fs.conf:

WTF?  In none of my examples did block size ever change.  In none of my 
examples was block size ever mentioned as a factor.

Inode density was demonstrably different in the resize vs. mkfs cases.

And online resize -obviously- imposes a limit on inode density, by 
locking inodes-per-group at fs creation time.  Dynamic allocation of 
inode tables would permit dynamic sizing of inode tables based on 
current needs, rather than needs determined at fs creation time.

	Jeff