Extent merging past MAXEXTLEN

Extent merging past MAXEXTLEN

[Michael Nishimoto]

Hi everyone,

Is there a reason to continue limiting extent merges past MAXEXTLEN
on 64-bit systems?

thanks,

    Mike

Re: Extent merging past MAXEXTLEN

[David Chinner]

On Mon, Jan 28, 2008 at 01:34:10PM -0800, Michael Nishimoto wrote:
> Hi everyone,
> 
> Is there a reason to continue limiting extent merges past MAXEXTLEN
> on 64-bit systems?

I don't think 32/64bit issues enter into this - the max extent length is
determined by the filesystem block size (the on disk length is in filesystem
blocks) so by default we are already at byte lengths greater than what fits in
a 32bit variable (i.e. 2^21*2^12 = 2^33).

Basically, MAXEXTLEN defines the number of bits in the length field of the on-disk
extent record when it is packed and so without a on-disk format change, we can't
sanely increase the size of this field. Perhaps we should look at doing this,
but it's going to have to wait until the btree re-factoring code is completed
so we can implement the record format change sanely....

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group