[Mageia-discuss] Mirror

[Michael Scherer]

Le mercredi 09 mars 2011 à 13:34 +0100, Juergen Harms a écrit :
> I have now arranged for a mirror to be set up at CUI - my old lab at the 
> university of Geneva - I hope it will come up in time for alpha-2.

That sound great, thanks.

> CUI is glad to help, but that is not a permanent solution - CUI normally 
> does not provide this kind of service - permanent mirroring services 
> should be provided by Switch as soon as a stable Mageia release becomes 
> available. I will pursue the discussion with Switch.

Ie, not permanent, do you know how and why this would stop ?

> Talking to people at Switch brought up some facts which Mandriva should 
> be aware of:
>   - There is a heritage of bad experience that has been made with Mandriva.
>   - It is not the first time that I hear the argument Mageia = Mandriva 
> = forgetit.
>   - Mageia happens at present to be perceived as "just another one of 
> those distros that appear and disappear to oblivion" (maybe with a 
> component of unfriendlyness towards Linux).

IMHO, our best bet for such case is to wait, so people can make their
minds and reconsider their position. I am not sure that trying to push
now is good. It take time to recover trust and make people forget where
do we come from.

Our problem is that if we say "we are a mandriva fork", people think of
their bad experiences. If we don't, we appear as just yet another
distro.

So maybe it would be better to engage first with people that do not have
negative experience with Mandriva rather than trying to convince people
who did have ?

> By the way, at my university there was a similar experience: Mandriva 
> was part of the officially supported infrastructure (mirror, 
> consultancy), with fees paid to Mandriva - that broke in dysharmony due 
> to bad administrative response from Mandriva, leaving quite some ill 
> feeling. Sorry if I wade through these negative arguments, but these 
> beyond-the-enthusiast-user spotlights count.

Well, we should make clear to people that we started the project for the
same reasons that they broke from Mandriva, and that we are aware of the
problems. But IMHO, we should not try too hard to convince them to help
us, just say that we agree with them. The rest will come by itself.

> I think that, when the stable release approaches, a small campaign to 
> rectify these prejudices would be an excellent thing.

That sound like a delicate task ( for the aforementioned reasons of
PR ). How would you start ?

> Talking with the people at Switch who maintain the Switch mirroring 
> service, there were also some concrete and technical arguments - I 
> quote, translating from German:
>   "the communication with the mirror sites at Mandriva had already been 
> minimal, compared with other distros: announcing releases, checking the 
> availability at the mirrors (QA), etc" (which is a mere quote, dont ask 
> me to interpret).
>
> Switch is reluctant to maintain a mirror at "assembly language level" 
> ("just run rsync every 2 hours"), they would prefer solutions using 
> something like MirrorBrain - but probably Mandriva experiences are part 
> of the background to this argument.

Well, what do they need more precisely ?

We understand well that they do not want to micromanage the mirror ( I
think no one does ), but then they can for sure understand that we
cannot adapt to everybody too on our side ( ie, another kind of micro
management ). 

And for mirrorbrain, I do not see how this relate to rsync, as the
software is used on distro side to distribute mirrors lists, if I
understand well. Rsync is still used to mirror around.

If I remember, the problem of switch was "there is not enough users to
justify the mirroring" ( as you explained before ), and I do not think
this will be solved until one or two years, nor that mirrorbrain will
produce more users by magic.

So maybe we should just wait to be credible when asking something to
them and be honest about that ?

> At present, putting a mirror on a univeristy site puts it into an 
> environment which is in good match to the straightforward rsync approach 
> - correct for the alpha period of Mageia.

Given the high number of university for all others distributions, I
think this is a good match for most of them :)

> But I think that on the long-run there are lessons to be learned from 
> the discussion with Switch - mirrors for a stable Mageia should be 
> preferably hosted at professionally run mirror sites (who make the kind 
> of consideration I quoted above), university solutions should come as 
> additional icing. In case of a second round of discussion on how to 
> organise Mageia mirrors, it might be a good idea to have that discussion 
> with some participation from the mirror sites.

I do not understand what you mean by "professionally run mirror". I 
think this doesn't mean what you want it to mean.

Switch.ch core "business" is not to run mirrors, this is to sell
connectivity ( to swiss schools ). The core "business" of a university
is not to host mirrors, but researchers and students.

>From my point of view, there is 2 motivations ( that can be
overlapping ): 
- people who run mirrors to help free software
- people who run mirrors because this help them on various level

A mirror can help to save bandwidth ( for example, free.fr is a french
provider, all linux user will likely go on their mirror as this is
faster ), or to leverage this for peering/trafic negotiation ( see
http://arstechnica.com/old/content/2008/09/peering-and-transit.ars , I
also recommend to see http://www.blogg.ch/uploads/peering-playbook.pdf ,
given at the end of the article ). 

I think most universities does this to help free software, and because
they use internally. I think most volunteer-run mirrors, like the one of
Wobo fall in this category too.

On the contrary, I suppose that most operators ( like switch to some
extend, but for free.fr, belgacom, etc ) does this for peering/trafic
reasons, and that mean they need to have enough traffic for this to be
useful for them, and enough users to justify the need. For now, I do not
think we are in a position to be interesting to this kind of mirrors.

So what we could do is to ask for some metrics so we could now when we
will be able to reach a agreement, and try to have a multiple step
plan :
- first, we try to engage some universities ( or likely minded admins,
such as kernel.org, or others ), to cope with the load of the
distribution. Some being 4/5. 

- based on the feedback, we make sure that the documentation is correct,
that our procedure works, etc, etc.

- once we have more and more people ( with metrics that have to be
found ), we try to find more mirrors.

- once we are interesting enough to engage people in the connectivity
business, then we try to engage them ( likely not before one or two
years, maybe more ).

-- 
Michael Scherer