On Thursday 18 of December 2014 10:00:35 Michele Calgaro wrote:
> On 2014/12/17 11:09 PM, Timothy Pearson wrote:
> >>>> Give it a couple weeks to mirror. Right now we are having
> >>>> some issues with our primary mirror (see the threads by Mike
> >>>> Bird on this list), and the financial state of the project
> >>>> cannot support a high bandwidth to upload them faster.
> This is not a "polemic" mail. Just want to raise awareness.
> I have been trying to test updating from 126.96.36.199 to R14.0.0 over the
> last 3 days, but the package download speed has been *ridiculously*
> slow for several big packages (like tdelibs for example), something
> like 821 B/s or 1109 B/s just to give some numbers. Not to mention the
> amount of timeouts I received.
> I understand the financial situation of the project (so the absence of
> any type of polemic complaint), but we are definitely not doing any
> good to us with this situation. Potential new users may just turn away
> when they try to download Trinity.
> IMO, in the future we must make sure that a package-wise fully updated
> mirror system is working properly before announcing the next release.
From my perspective: Slow line on the primary server is a fact about which
there is no need to discuss. I dare say that this slow line may not be a
fatal problem. But we would have had to draw, how to effectively use it.
I see several problems:
1) The more mirror sites that sync from the master server => the line is more
overloaded => synchronization is complete in sight.
2) The longer synchronization time to mirrors causes that more users download
files from the master server => the line is more and more overloaded =>
everything is moving towards infinity.
3) Nightly-builds and preliminary-stable-builds PPAs are not on the mirrors =>
their users cause additional overload of the line.
As we can see now, every additional user deepens this problem, each additional
user is frustrated. I believe that we need to prevent any user to access
directly to the master server.
I propose to establish an official primary mirror, which will be on enough
bandwidth and will be the only one who will access to the master server. All
other mirrors would then be synchronized from this primary mirror. This would
solve the problem 1), including the consequent problem 2). On this primary
mirror should also be mirrored nightly-builds and preliminary-stable-builds,
to avoid also the problem 3).
What is your opinion?