[reiserfs-list] Re: [OpenAFS-devel] more on the 2.2.18pre17 SMPcpu hog/etc.

Derek Atkins warlord@MIT.EDU
04 Dec 2000 19:14:27 -0500

Hans Reiser <reiser@namesys.com> writes:

> but even having 64bit support is not really good as a long term
> architecture for anything that requires a lookup from the FS.
> Basically, unless you have variable length keys, you cannot have
> stem compression of directory entries and all sorts of other nifty
> things.  filenames are a much cleaner interface.

I think this argument is going far beyond AFS, and should probably
move off the AFS list.

> objectids (inode numbers are objectids) are what I once thought was
> a good feature, and then realized were badness.  You cannot help but
> destroy locality of reference with objectids.

Huh?  Just build the locality information into your objectID.  For
example, in a 64-bit inode, use the upper 32-bits as the locality and
the lower 32-bits as the ID within that locality.  I think the problem
is that you are assuming an objectID must be random, but that is far
from the case.

> hans

       Derek Atkins, SB '93 MIT EE, SM '95 MIT Media Laboratory
       Member, MIT Student Information Processing Board  (SIPB)
       URL: http://web.mit.edu/warlord/      PP-ASEL      N1NWH
       warlord@MIT.EDU                        PGP key available