Re: Variable length fields

[James Rich <james@xxxxxxxxxxx>]

On Thu, 1 Jul 2004, James H H Lampert wrote:

> Personally, though, I find it hard to believe that the VARLEN space in a
> file NEVER gets garbage-collected or burped. After all, Java
> garbage-collects its heap rather frequently, on much less efficient boxes,
> and what is the VARLEN space but a heap that just happens to reside on a
> disk, instead of in memory? Then, too, the hard drives of DOS boxes,
> Win-Doze boxes, (presumably) Linux boxes, and Macintoshes get defragged
> every day, and what is that, but a variation on garbage collection and
> file-burping?

And that presumption is incorrect.  Defragmentation on linux is relatively
uncommon or, depending on file system, non-existent.  Not because no such
tools exist but because the filesystems are smart enough to not need it.

However, on linux (as most other OSes) the filesystem is not the database.
Usually the database exists on top of the filesystem, but is not an
integral part of it (the exception here being databases such as Oracle
that can use the raw device and create their own filesystems on it and
manage it themselves).  Thus any defragmentation tools wouldn't
necessarily affect database file sizes (they still could, just not
necessarily).  What would be interesting is to see how a database such as
MySQL allocates space for it's TEXT data type and what happens to that
space when a record is deleted or modified with shorter contents.

James Rich

Vs lbh cynl n Zvpebfsg PQ  onpxjneqf, lbh pna urne fngnavp zrffntrf. Ohg
rira jbefr, vs lbh cynl vg sbejneq, vg vafgnyyf gurve fbsgjner!
        -- Fcbgvphf ba /.