[CVALE] The Verdict is.... ( was Re: hans reiser on 20/20)

[Pete Zaitcev]

On Tue, 29 Apr 2008 08:09:34 -0700, "Landon Blake" <lblake at ksninc.com> wrote:

> Thanks for posting that story Patrick. It was very interesting.
> 
> Where does this leave his file format?

Rotting outside of the Linus tree.

Look, it's not like reiserfs was all that in the first place.
The idea to layer the filesystem on top of an object database
is very attractive because of its performance implications.
You get constant creation times, instant delete. Data transfer,
however, is governed by the design of the underlying database
(e.g. if it has extents). And in bargain you receive a terrible
reliability. One bad block in the right place, and huge chunks
of the file system become inaccessible, because it has no
redundancy of any kind. Surely it's something the database
can address, by adding complexity to it, but then the whole
thing just balloons without end. In present, using reiser3
or reiser4 is a recipy to lose your data, including read-only
data. SuSE learned it the hard way.

Also, to extract maximum value quickly, Hans went the route
of exposing the database through the VFS to applications.
The result was a hideous mess which doomed reiser4 as a project
regardless of its "technology" (e.g. the architecture and
the implementation).

In the end it turned out that traditional filesystems were not
quite out of steam yet, as btrfs demonstrates.

-- Pete