[parted-devel] [PATCH 06/11] libparted: respect optimal IO alignment

[Phillip Susi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 01/07/2013 10:53 PM, John Gilmore wrote:
> I think this patch is problematic.  Merely repeat-posting it to the
> list does not make it any better.

It is, but only as I stated ~9 hours ago, it is because I
misunderstood the code.

> If you're creating a RAID array with an odd number of drives, and 
> partitioning it, set your partition starts as sector addresses (as 
> opposed to values like the default or "0" or "1m"), and parted
> won't mess with them.  That's the only situation in which the 1mib
> default produces bad results.

This would not have been an acceptable solution, since parted would
have still complained about it being misaligned, and, more
importantly, requires manually selecting exact aligned start/length in
sectors to get the correct results, which are supposed to be picked
automatically.

As it stands, the code goes with 1 MiB even though a smaller power of
two would produce optimal throughput but with less wasted space.
Since the only loss is a small amount of wasted space, that is an
acceptable tradeoff for a slightly more portable partition table.  I
had thought it would chose 1 MiB over any non power of two size, such
as 768k, or 1.5 MiB, which would be a problem as it would result in
bad throughput as well as some wasted space.


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with undefined - http://www.enigmail.net/

iQEcBAEBAgAGBQJQ654bAAoJEJrBOlT6nu75R0YH/3c1CP2S3pIg/eQQCU9B3jnU
j5FunZdWYptBzeCjh3YbPE1NAViD5zwuBqElTl+aWJxBFZhYX4lT3WAKexelTb73
JGGl48Y7jH9O5A++1KQDpLzSNBfOQjMaO7Asvujx2kztBi+ZQY6WJl50G/glcwg6
dHx87jHIfcRyf72AxQGTtjs8blxH12iyAqUVP/FggRQebBLuj74xcN72sQADFh/j
nh1y+vezdD23bs+SzG8NzwGfyLselxTCxwme2LHwVxOBWL9ySCZ6lLrobWs9IXhU
agARhx8XxW55l97kFkrVh092LbDURa/jM5wHSWF5p6Bem3EqSAwrs8i+pGARsMw=
=uruU
-----END PGP SIGNATURE-----