On the conviction of Hans Reiser [LWN.net]

On the conviction of Hans Reiser

Posted May 1, 2008 9:15 UTC (Thu) by jospoortvliet (subscriber, #33164)
In reply to: On the conviction of Hans Reiser by walters
Parent article: On the conviction of Hans Reiser

just out of curiosity: aren't there any advantages to having database-like 
capabilities in the filesystem layer? Looking at what beagle, strigi and 
tracker are doing, or even more interesting Nepomuk? I wonder if it would 
be more efficient to have that stuff in the filesystem like BEOS had?

(Log in to post comments)

On the conviction of Hans Reiser

Posted May 1, 2008 9:48 UTC (Thu) by MathFox (guest, #6104) [Link]

just out of curiosity: aren't there any advantages to having database-like capabilities in the filesystem layer?

There are advantages in having search and annotation capabilities in a file system. There are significant disadvantages too.

Keeping search indices up to date takes CPU time and memory and requires additional disk access above the ordinary writing of the file. Furthermore the search index eats up disk space. Writing a text file to a "database-like" file system will be slower than writing to an efficient POSIX style file system.

The biggest issue there is is that nobody knows what features should be supported by the new file systems... It is very likely that it will be different from POSIX, but the kernel developers don't implement hand-waving. It makes more sense to build some prototype "storage daemons" and demonstration applications to get a feeling about how the interface should look like. It is an interesting research topic, but not ready for a stable production OS yet.

On the conviction of Hans Reiser

Posted May 1, 2008 13:23 UTC (Thu) by rwmj (guest, #5474) [Link]

It makes more sense to build some prototype "storage daemons" and demonstration applications to get a feeling about how the interface should look like. It is an interesting research topic, but not ready for a stable production OS yet.

Since you ask:

ZipperFS

The talk is quite interesting, and he did implement it (albeit as a telnet daemon rather than an actual filesystem). I believe there is an NFS-daemon implementation of this too but I can't find it right at the moment.

Rich.