Re: Speeding Up APT - Project Idea
On Fri, Feb 4, 2011 at 17:28, Ishan Jayawardena <email@example.com> wrote:
> I would like to do a project under APT for this year's summer of code
You are welcomed. :)
I was a gsoc-student last year, so feel ask if you have questions.
I would recommend to join IRC - #debian-apt and #debian-soc will
be interesting for you to get contact to the right people.
(sidenote: If my homeserver isn't broken i am idling on both channels)
> installation for apt/dpkg and multicast apt. I also wrote to the
> debian-dpkg list about my interests and received couple of feedback.
For reference, this thread starts here:
> They had mentioned some suggestions for speeding up the installation
> process of apt such as, the usage of self-contained batches of
> packages, doing parallel downloads, running dpkg in parallel with apt
This topic turns up in a recent thread on debian-devel as well 
so its a really interesting topic and could make really a good gsoc
project as it is not too hard as i fear the streaming stuff needs too
many changes in APT and dpkg (and properly other programs) to
be come true which will be hard to do in just ~3 months.
Splitting up downloading and installing into smaller chunks on the
other hand should need 'only' changes in APT and the current download
system is designed to download packages in the required order, so
you 'just' need to find the right position where you can start a
parallel install process…
> etc.. Also they had mentioned about the official adoption of debdeltas
> for Debian.
That is another interesting thing. The problem is currently that a download
method doesn't get to know which version of the package is currently
installed (if any) which is needed as far as i understand it to repack
a package to apply the debdiff to… but yeah, nothing which would be
impossible to solve i guess.
> Currently I am learning the internals of APT and looking
I think the best way to get to know your way around is to tackle some
bugs - the good thing is that you have a lot to choose from¹ - which also
helps to prove to the gsoc admins that you know what you are talking about
which increases your changes to be accepted into the program.
(just as a general hint)
¹ the first time i consider this a good thing… ;)
> would like to know your ideas and suggestions about these projects and
> other new project ideas that you would have related to APT and package
Good question! Zero-bug-count would be cool²… ;)
More seriously, i thought already in my proposal last year about a way
to exclude packages from being included in the binary cache which could
be used e.g. on servers to exclude gnome and kde packages from
even be parsed into the caches which speeds up the dependency
checking which can be quiet noticeable on slower devices.
Can't think about other gsoc-fitting projects now, but its pretty late in
my timezone now, so maybe after the weekend…
² just said to fix the balance in the universe again after ¹
P.S.: I noticed that you were cc'ed to all mails on dpkg list, so i cc'ed
you here, too, but the normal case is to not do this if not requested
by the sender -- see code of conduct of debian lists.