> Would there be any interest in starting a project to
> deliver indexed
> attribute searching functionality on top of ZFS, akin
> to WinFS or Be's
> BFS file systems?  I've not yet researched the
> IP/patent implications
> of such functionality, so I'm perfectly willing to
> hear that it's a
> dangerous notion.
> 
> Are there any internal projects underway with similar
> functionality by
> chance?  Have others considered this and ruled it
> out?  I can expound
> further on my thoughts if anyone is interested.
> 
> Comments?

Are there even any _external_ (Linux, *BSD, proprietary Unixes)
applications that take advantage of or allow the user to interact with
such capabilities?  Yes, there's e.g. Windows XP's Indexing Service,
and Mac OS X's Spotlight, and and they probably do represent the bulk
of most user's experience/expectations between them.  But there are
some problems, IMO.  Read http://www.nasconf.com/pres04/waidhofer.pdf
just to get some ideas about the filesystem itself, first; note the division
between lightweight "Extended Attributes" that may be significant to the
OS, and basically subfiles or named streams, which are strictly application
territory.  Then there's the problem of standard names for things: without
those, how the heck does everyone agree that a particularly named
Extended Attribute contains for example a MIME type, or that a particular
named subfile contains a thumbnail (not to mention the preferred format
and dimensions of the thumbnail...).

Now I think I've seen some talk about intercepting file operations, and
there's also the user-space FEM API; either or both might be handy giving
an indexer the hooks needed to maintain its index, once it's initially built it.

But (aside from perhaps a clearer distinction between what other
systems mean by Extended Attributes vs subfiles or named streams), I'm
not sure I understand what you what zfs itself to do that it doesn't already.

Also, a Really Cool indexing service needs some agreement on a
schema for useful metadata naming, as do fancy file managers
(and applications that know something of the intent of the files they create),
so that the index isn't just a pile of unstructured garbage, but rather 
something
smart enough to find all audio/mpeg files (without resort to extension or
magic numbers) and scan their ID3 tags for particular artists; or all ODF
files with certain content, or ...  Seems to me the _first_ thing that
everyone needs to agree on is what name and interface to use to
store or retrieve the MIME type in (as proper metadata, not something
deduced by filename extension or a typing engine)!  But darn if I can find any
evidence that there's anything like a standard that far along...
 
 
This message posted from opensolaris.org
_______________________________________________
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss

Reply via email to