[Gnu-arch-users] Re: Avoiding ancestor scan during get with revision library

[Stefan Monnier]

>>> tla directory: 4.4 M, tla.tar.gz: 203 k
>> And then try bzip2 to get even more disgusted,
> On similar data, 65k.  That's about 69:1.  bzip2 takes significantly
> longer than gzip, though.

Compression time is not important here.  Decompression OTOH needs to be fast.

> I suspect the whole archive format would have to go bzip2 for bzip2 to be
> tolerated, and you'd probably want to compress a smaller unit, like
> a given package or package-version.

Fair.  And gzip compression already gives pretty good results.
BTW, log compression would probably be done a branch at a time, so the
above compression result (where all the branches are compressed together)
is a bit optimistic.

> On Reiser, it's in between, at 3.5M

I don't care that much whether it's 3.5M or 4.4M: the access time
is what counts.  Note that if the files are sufficiently poorly layed
out, it's quite possible that unzipping the compressed file is faster
than catting the uncompressed files.


        Stefan