Linux Registry, not only the issues side

George jirka at 5z.com
Sun Apr 18 09:11:02 EEST 2004


On Sat, Apr 17, 2004 at 12:32:54PM -0300, Avi Alkalay wrote:
> Some may think that one file per key will consume many filesystem
> i-nodes. Actually, when not using Reiser4 filesystem, it may consume some
> more disk space, and it may also be not so efficient than reading one
> single text file, as KConfig does. But Registry's nature let applications
> load their keys on demand; so it is possible to avoid the
> read-all-use-some approach. Writing updated keys back to disk is also
> more robust, because unchanged keys won't be touched, different from a
> single file approach, that must be entirelly rewritten.

Note that read-all-use-some can also be avoided in say an .ini style format
by just not keeping things in memory and just reading line by line until you
hit the key you want.  And this is in terms of algorithms the same as
traversing a directory structure on most filesystems (unless you use some
btree filesystem or some such).  And will use theoretically the same amount
of memory.  On the other hand reading lots of small files will completely
whack out the filesystem cache making some other things having to reread
things from disk.  Most filesystems will cache a single file FAR better then
many small files.  So many small files will in fact be a big hit on the cache
and on memory unless the filesystem is tuned for this.

> Besides that, big applications (like Mozilla, Konqueror, KDE, Gnome) key
> gathering time is a very small part of what they have to do to start up.
> And the benefits of an homogeneous registry to all system are much bigger
> then these issues. Think about a common integration between everything,
> flexibility, security granularity and openness.

Besides using buzzwords what are the actual advantages of this?  If the only
advantage is standartization why not use some implementation that doesn't
have those efficency issues?

George

-- 
George <jirka at 5z.com>
   The ability to quote is a serviceable substitute for wit.
                       -- W. Somerset Maugham




More information about the xdg mailing list