Bandwidth problems at beta time (apt-get is slow)

[Martin Olsson]

Jono Bacon wrote:
> On 10/02/2009 09:05 PM, Jorge O. Castro wrote:
>> Ken asked me to help him test some empathy stuff today and grabbing
>> the build-deps took over 10 minutes for a few hundred k's worth of
>> packages. I have a caching proxy at home so normally it's not so bad
>> but I am starting to wonder if there should be an ACLable mirror
>> available to people in core-dev/MOTU/bugcontrol/whatever so people can
>> get the important stuff done. In the past I've gotten emails from the
>> ISO tracker before my original rsync is even close to done so the pain
>> just multiplies over the course of the end of the cycle..
>>
>> I think it's great that we have so much enthusiam for a release but
>> surely we can set aside a certain percentage of bandwidth for people
>> to get their testing done.
>>
>>    
> I agree. I think we need to find a better solution for this problem. 
> What does everyone else think?

I was also frustrated by this yesterday while doing upstream
development I needed some -dbgsym packages to hunt down a bug.

Any solution that treats core-dev/MOTU like a special case would
be a big mistake because A) lots of upstream devs use Ubuntu, and
B) the bandwidth issues are also a _huge_ turn off for normal users
installing or upgrading Ubuntu.

The real solutions include:

A) use bittorrent to download .DEBs - this is not very hard to do
when you think of it; we got many high quality bittorrent client
libraries available already. If this is hard for some reason, I'd
love to hear why; maybe we can think of some way to simply the
problem?


B) dont send full DEB's all the time, send a bsdiff on upgrade,
like how Google Chrome distributes their updates.


C) create more mirrors and make sure each user can download
for multiple mirrors at once (i.e. do load balancing between
mirrors, dont "hardcode" a specific mirror in preferences).


D) upgrade the servers and their internet connections



Personally, I think A) provides the best benefit / cost ratio.




		Martin