Re: Speeding Up APT - Project Idea
Hi,
I wrote a project proposal[1] and inserted it into the idea's wiki
page[2]. Can you give me your feedbacks on it?
Thanks.
[1] http://wiki.debian.org/IshanJayawardena
[2] http://wiki.debian.org/SummerOfCode2011/AptDebdeltaIntegration
On 3/26/11, Michael Vogt <mvo@debian.org> wrote:
> On Wed, Mar 23, 2011 at 11:27:48AM +0100, Goswin von Brederlow wrote:
>> Ishan Jayawardena <udeshike@gmail.com> writes:
>>
>> > Hi Michael, Hi all,
>> >
>> > As you may remember, I talked with you about integrating debdelta into
>> > apt-get, on the IRC. I found it interesting and therfore, hope to come
>> > up with a project proposal out of it.
>>
>> It might make more sense to look into zsync and implement the missing
>> support to "look into .deb files". The .zsync files could then be
>> generated by DAK and could be used with any http mirror.
>
> The debdelta integration is a good solution for stable and for
> security updates IMO. We know it works well and its stable since some
> time and well understood. But not that great for unstable. For known
> version updates (like security updates) it should produce smaller
> deltas than zsync too.
>
>> The advantage of zsync would be that it can download the delta between
>> any 2 versions of a package or between a damaged and actual file of the
>> package. With debdelta the mirror would have to contain the exact diff
>> between the local version in the apt cache and the current version. That
>> would be quite unworkable for unstable. Ad the deltas would be far
>> larger than the .zsync files.
>
> The zsync based delta would indeed be preferable (if we can make it
> work!). There is even working code around around it here:
> https://code.launchpad.net/~felix-feyertag/apt-sync/main
> But we never quite finished it and it needs someone to really look
> into it again and see whats missing. More research around it is
> needed, like how well does the deltas work, what blocksizes to use,
> the http range request approach may turn out to be problematic with
> various proxies etc. I would love to see someone working on it too,
> but I consider it a bit more experimental then the debdelta approach.
>
> Cheers,
> Michael
>
>
--
Regards,
Ishan Jayawardena.
Reply to: