On Thu, 2006-09-07 at 13:08 -0500, Johnny Hughes wrote: > > > CentOS 3 wont have the same issue because all updates are referenced via a > > > round robin set of mirror.centos.org, which will always be seen as the > > > same url by the proxy. > > > > What's the problem with that scheme? It's hundreds of times faster > > on my second and subsequent machines - and would be for anyone else > > going through a proxy configured to cache large objects. > > What is wrong with that scheme is that only 1 mirror is listed ... if > you loose the connection, if it gets overloaded in the middle of your > transfer, etc. then there is no failover. Doesn't your geo-ip enabled DNS service drop non-responding servers? It has been much less trouble in practice from my locations than the fedora or centos4 repositories. > > What I want is for the default install to work using standard > > cache techniques. The 'you can configure' concept only works if > > you know everyone else sharing the same proxy and can pre-arrange > > every file download with all of them which is pretty unlikely in > > any organization large enough to even have a proxy. A scheme that > > uses the same URL for the same file will always work automatically. > > If you do this, spreading the load won't be an issue since there > > will actually only be one download. If you can't use the same > > set of mirrors as centos3, there has to be some way to make this > > happen based on some computation that would be repeatable - perhaps > > a server-side check that can see the requester's public source > > address and give back the best mirror URL or a sorted list that > > would always be the same for that source IP. > > If you have alot of machines doing the same updates from the same place > (and so are behind a proxy server) then hosting the updates yourself on > a local mirror would be much easier and you would not have to worry > about any of these issues. I don't see how that helps with different people who don't spend all their time coordinating file downloads with each other. And you'd have to repeat the work for every distribution and every repository and for every installed machine - and at every location. I don't see the 'much easier' part anywhere in this picture. How is anything 'much easier' than having the default system work the way caches have been designed for at least a decade? -- Les Mikesell lesmikesell at gmail.com