Le jeudi 17 février 2011 à 11:57 +0100, Juergen Harms a écrit : > There is some history about this issue, worth knowing. > > For many years, Switch had hosted a Mandriva mirror. Sometime, about a > year ago, that mirror disappeared. I had then contacted Switch to find > out what had happened (that was before the time of Mageia). > > The reply was that the usage of the Mandriva mirror had dropped to such > a low level, that Switch had decided to remove the mirror - the list of > software mirrored at Switch is quite remarkable, but without some > housekeeping the amount of mirrored data gets out of control. I had this > discussion with > Thomas Lenggenhager <[email protected]> > (Thomas is the head of the group responsible for the mirror). > > Later, when Mageia had started to prepare its release, I had a > discussion with Christoph Graf - the vice-director for Network services > - and asked him about the policy of Switch with regards to adding a > mirror for a new Linux distribution - which, necessarily, starts from a > close-to-0 usage level. We agreed that Switch feels it its mission to > give new distributions a chance. At that time we did not press the > discussion further, waiting for the first Mageia (pre-)release to become > available. > > I agree with you that Switch is an ideal mirror for hosting a Mageia > mirror, I suggest you send the information pointed to by Michael to (but > only when information relative to item (5) of the howto exists - if > switch does not have the information how to register, chances are big > that the request ends up in some todo-sometime-later list) > > [email protected] > > I have read the howto at > http://distrib-coffee.ipsl.jussieu.fr/pub/linux/Mageia/mirror.readme > I have corrected some typos (it is a short document, I append the > corrected text to this mail). Regarding the objective contents, I have 3 > comments
Thanks, I have updated it. ( and will need to add other corrections ) > (1) is a 2-hour interval for syncing not exagerated? I think to remember > that with Mandriva there was one sync during the night; a longer period > reduces the load on the server, and reduces the periods during which > inconsistencies might be possible; probably a distinction between > cauldron (tier-1) and production (tier-2) requirements should be made > (coming back to Switch: Switch did not host cooker). I think the instruction for now are only for cauldron, since that's the only thing we have so far. But yes, maybe for stable release, this should be changed. > (2) is it wise - at the tier-2 level - to suppress checksum protection? > I would say that with the relatively low number of tier-2 servers the > additional load on the server should not be prohibitive (particularly if > the interval of syncs is made longer). Mistakes due to transmision > errors at this level have a high multiplication rate and a substantial > nuissance potential - and certainly do not help to improve the image of > the distribution. Again, there is a difference in what is tolerable in a > development and in a production environment. Well, in fact, given the way tcp work, I do not see how it could suffer from a transmission problem. I mean there is checksum, we are not using udp to transmit anything ? > (3) should the form not also specify the email and telephone number of a > contact person to be contacted in case of problems? A email for sure. A phone number is less useful ( ie, you need to speak english, to call on long distance ). That's just a mirror, after all. For tier1, that's a different story of course. -- Michael Scherer
