Re: mirrorlist server selection

On 3/17/2010 3:54 PM, Seth Vidal wrote:
>
>
> On Wed, 17 Mar 2010, Les Mikesell wrote:
>
>>
>> I don't think the underlying connect() routine will do it for you, but
>> browsers and their associated libraries must do it the hard way. Try
>> putting a bad and good IP address in DNS for a name and using an
>> assortment of browsers, wget, curl, etc. to access a file at a url
>> that works on the good IP. With clients that know how to to this,
>> round-robin DNS is a cheap and moderately good failover mechanism as
>> well as providing load balancing among the sites that are working.
>>
>
> Understood - it will require a little bit of patching to implement this
> - and it kinda defeats the point.
>
> Seems like with yum's mirrorlist and metalinks functionality you just
> list the individual sites that you have listed in your dns records and
> it'll just work, the same.

So you have to maintain 2 things in sync instead of just one, and since 
every host behind a cache is likely to choose a different mirror name 
you end up downloading 10 copies of every update - or however many 
mirrors there are.

-- 
   Les Mikesell
    [email protected]