[lug] Microsoft invents file links

[Archer Sully]

----- Original Message -----
From: "Judson Valeski" <valeski at netscape.com>


> Archer Sully wrote:
>
> > This one was covered to death on Slashdot.  Its a little bit more than
> > symlinks.  Its an automated system that scans the file system looking
> > for identical files, which are replaced by a symlink to a file in a
central
> > location.   I am curious to hear how MSFT have worked around the
> > many obvious holes in such a scheme.
>
> hmm. If this is handled at the file system level, I don't see the holes.
Seems
> to me you could simply reference count those "using" the file. "Using"
seems
> loose... let's see...
>
> Unless they went wild with access caching, nah, that would never work...
>
> The only *real* "users" of a file would be those who create it, and those
who
> delete it. If you use that as your rule for reference counting, this seems
> like a pretty straight-forward and effective optimization.
>

For user data it is an ineffective and potentially harmful optimization.
The
problem is that if file identity is based solely file contents, then you
need
some sort of COW semantics for the file system, something that is not
mentioned in MSFT's paper.

What it really appears to be aimed at are the multiple copies of various
applications that get installed on everyone's machine in a Windows
environment.  And this only a problem because Windows is so ineffective
at sharing resources.

-- archer