Web lists-archives.com

Evaluation (Re: Proposal: A new approach to differential debs)




On Sat, Aug 12, 2017 at 02:16:21PM -0400, Julian Andres Klode wrote:
> Hi everyone,
> 
> (I CCed -devel and deity, but we probably should just discuss
>  that on -dpkg)
> 
> while breakfast here at DebConf, the topic of delta upgrades
> came up. I think delta debs are generally a thing we should
> aim to have, but debdelta is not the implementation we want:
> 
> * It is not integrated into APT or dpkg
> * It relies on a script shipped in the debdelta to generate
>   a new deb from and old deb

> We guessed that generating the new deb from the old deb is
> actually the slowest part of the whole debdelta process. I
> propose a new solution, which does not really have a name
> yet.

# Evaluation

I build a few sample delta debs, using bsdiff (.pdeb) and
xdelta3 (.xdeb). I attached the scripts to generate them
and apply them to an unpacked directory tree.

Feel free to check with other packages, here are the current
evaluations, including a comparison against debdelta.

libreoffice-core (size only):

  -rw-r--r-- 1 jak jak 29M Jul 22 20:02 libreoffice-core_5.3.5~rc1-3_amd64.deb
  -rw-r--r-- 1 jak jak 31M Jul 16 00:10 libreoffice-core_5.4.0~rc2-1_amd64.deb
  -rw-r--r-- 1 jak jak 31M Jul 28 18:29 libreoffice-core_5.4.0-1_amd64.deb
  -rw-r--r-- 1 jak jak  18M Aug 15 23:44 libreoffice-core_5.3.5~rc1-3_5.4.0-1_amd64.pdeb
  -rw-r--r-- 1 jak jak 4.5M Aug 15 23:42 libreoffice-core_5.4.0~rc2-1_5.4.0-1_amd64.pdeb

For 5.4~rc2 to 5.4 it made a huge difference, for 5.3.5~rc1 to 5.4 not so much,
so it probably is a good fit for stable updates, but not for unstable and testing.

firefox (size & performance):

 -rw-r--r-- 1 jak jak 2.3M Aug 15 20:59 firefox_55.0-1_55.0-2_amd64.debdelta
 -rw-r--r-- 1 jak jak 2.4M Aug 15 22:13 firefox_55.0-1_55.0-2_amd64.pdeb
 -rw-r--r-- 1 jak jak 7.4M Aug 15 22:36 firefox_55.0-1_55.0-2_amd64.xdeb
 -rw-r--r-- 1 jak jak  38M Aug 10 06:49 firefox_55.0-1_amd64.deb
 -rw-r--r-- 1 jak jak  38M Aug 10 12:44 firefox_55.0-2_amd64.deb

Generating the -2 deb from the -1 deb using debdelta took about 47 seconds. In
contrast, applying the .pdeb and .xdeb files to an installed directory tree
took about 1.5 seconds.

The .pdeb uses bsdiff compression, the .xdeb uses xdelta 3. It took
96 seconds to generate the pdeb, and 13 seconds to generate the xdeb
on my ThinkPad X230 with 16 GB of RAM and a Core i5-3320M.

# Conclusions

1. xdelta3 is substantially faster than bsdiff, but bsdiff produces substantially
   smaller update sizes.
2. deltas for major updates are too big to be massively useful, so focusing on
   stable and security updates seems like a good idea (though we do need to have
   a set of pdebs for testing in unstable and/or testing)

# Further extensions

If you put a pristine-tar delta into the delta file, you can fully
reconstruct debs.

-- 
Debian Developer - deb.li/jak | jak-linux.org - free software dev
                  |  Ubuntu Core Developer |
When replying, only quote what is necessary, and write each reply
directly below the part(s) it pertains to ('inline').  Thank you.

Attachment: build-patch.sh
Description: Bourne shell script

Attachment: apply-patch.sh
Description: Bourne shell script