Re: [Gnu-arch-users] archive storage format comments on the size

[Andrea Arcangeli]

On Mon, Sep 29, 2003 at 05:45:35PM -0500, Charles Duffy wrote:
> On Mon, 2003-09-29 at 17:39, Andrea Arcangeli wrote:
> > this isn't a big deal actually I measured that too. A gzip -d of the
> > 3M .tar.gz generated a 65M tar file, that means removing the overhead of
> > the filesystem, the uncompressed size was still over 65M. There was no
> > waste of space due the fs in the tar file.
> 
> IIRC, tar pads out its files to end on a 10k boundary, and has internal
> padding such that even a 1-byte files takes about 1k to store. Hence, I
> wouldn't quite go as far to say "no waste of space due to the fs" -- as
> long as you're willing to consider the tar format a filesystem for those
> purposes.
> 
> (yes, I'm nitpicking -- sorry).

I never notice the 1k alignment of files in tar in practice, sounds
like if they didn't want to store a 64bit for defining the i_size but I
never looked into the tar sources so I'm just guessing. Ironically some
fs does a better compacting job than tar. About the 10k boundary, that
one doesn't matter at all is a >65M file.

with say 4000 files that would be still only at most a 4M different more
or less for a 65M file, but it's very fair to point this tar detail out,
thanks for the info.

Andrea - If you prefer relying on open source software, check these links:
            rsync.kernel.org::pub/scm/linux/kernel/bkcvs/linux-2.[45]/
            http://www.cobite.com/cvsps/