Re: Packaging _still_ wasteful for many large packages
Joey Hess <email@example.com> writes:
> Goswin von Brederlow wrote:
> > The problem is that every byte not shared is multiplied by 11 (soon
> > 12), extra Packages on the other hand only add a few bytes to the
> > Packages file.
> More like a kilobyte per package, per Packages file. Yes, I have done
> the math, and it's not clear to me if a few kilobytes downloaded daily
> by many of our users, some on thin pipes, has a lesser cost than a few
> megabytes sitting in a few mirrors. Especially since bandwidth is
> generally more expensive than disk. That's why I asked where the dividing
> line is.
Wasn't a 10 or 5 MB line suggested?
One thing you are right about. Splitting up the packages adds 1K
(certainly less compressed) for every Packages/Sources download. The
data on the other hand just wastes mirror space and traffic when
mirroring (only once per update).
So 10000 people downloading a 1K bigger Packages file is roughly the
same as 1MB data not being -all. Or 1000 people downloading a 1K
bigger Packages file 10 times (in as many days for example).
But if you think about it you should advocate using rsync or diff to
update the Packages/Sources files from day to day. That saves far more
traffic already than the splitting up would add.
> If we do not do something, Debian may be completly unusable for dialup
> users within a few years. I can already only manage to update my
> unstable systems once a week. Is this important? More or less important
> than the number of mirrors we can field?
> > Even if that doubles the number of packages its not
> > realy more complex.
> And you go on to propose some complex ways of dealing with it in a UI..
I keep forgetting you are a poor modem user. Downloading a 20MB
Packages/Sources files instead of 10MB realy is a problem. If I were
you I would have switched over to rsyncing the Packages/Sources files
PS: rsync for the Packages/Sources files, not .gz or .bz2 and just for them.