[CentOS-devel] mirros - for the record
lance at centos.org
Mon Nov 19 12:53:08 UTC 2007
On Thu, 15 Nov 2007, Ralph Angenendt wrote:
> Karanbir Singh wrote:
>> David Hrbáÿÿ wrote:
>>> I have found this solution for mirrors redirecting, SUSE way. I'm
>>> sending it for the record, might be useful.
>> how does this compare with what we have in place at the moment ?
It seems to me that they may have copied what we do - like fedora have -
isnt there some maxinm that says that great things arent likely to have been invented
twice ?? :)
> Two differences: Mirrors are sticky (so I get redirected to the same
> mirror when I request another file within 30 minutes)
That would seem to be a lot of caching for little benefit IMHO.
> and the redirector
> checks if files are on a mirror *before* it redirects to the mirror
> (does it really do that or do I have some conceptual misunderstanding
Their system works in similar way to ours with a crawler, although it
would be possible to check prior to redirect - the request is just a HEAD
They use the system for general mirror file download requests, whereas we
only use it for iso downloads and to give out geographical
mirrorlists for mirrorlist requests.
Users can still browse the trees on mirror.centos.org and download
I can see that there may be a benefit to redirecting access to individual
package downloads, especially with large packages such as oOO.
> a) can be resolved by using the fastestmirror plugin.
Indeed, assuming user is using mirrorlists and not hardcoded
> b) seems to result in some more load on the mirrors.
Right - although minimal traffic - and results could be used to generate
live state data for the mirrors. Obviously if redirecting large numbers of
files it would be better for the check to be moved to the point of access
rather than the crawler - although I would like to see the results cached
for a period of time to minimize load on the mirrors.
The ISP of choice for the discerning Linux user.
More information about the CentOS-devel