[dns-operations] opting in to stupid DNS tricks

[Patrick W. Gilmore]

On Feb 21, 2011, at 18:57, Joseph S D Yao <jsdy at tux.org> wrote:
> On Mon, Feb 21, 2011 at 09:55:24AM -0500, Patrick W. Gilmore wrote:

> Taking my heart in my hand, I approach this slightly venomous dialogue,
> hoping that Patrick will be good enough not to attack me for questioning
> him.  And, yes, others used mean language first; but escalation is not
> the way to win or even wage a discussion.

I'll do my best.


> In fact, one of my "best practices" list is to have all name servers
> serve the same information.  I was thinking of people who hand-enter
> changes on all servers, and don't bother with such things as zone
> transfers, and inevitably get discrepancies.  I did add an exception for
> split DNS.  This points up another exception I should probably add,
> since we use devices that I've called "conditional" name servers that
> return different IP addresses under different conditions.

Mind if I ask what your question is that so worried you?

What you say above is exactly what Jim & Paul are claiming is bad.  While they may attack you as they have done Akamai & CDNs in general for admitting to such grievous sins, I shall not.


>> However, just to clear up a few things, CDNs are not the only companies that get geo-localization wrong.  And there is the fact most web page localizations (including all the Akamai ones I've seen) use the client IP address, not the name server IP address.  But please don't let things like facts get in the way of rant against CDNs.
>> 
> 
> I am curious about how this works.  As I have observed it, I thought
> that Akamai returned an IP address of a local mirror.  [At least, one of
> their services does.]  In that scenario, the end client's IP address is
> never seen: only that of the recursively resolving name server that
> queries Akamai.  I could understand it if the Web browser went to a
> central Web server that then "permanently" redirected the URL to a
> closer server - but that, again, is not how I understood Akamai to work.

You are conflating which web server serves the content with what content is served.

Yes, Akamai returns a different IP address for each query depending on your (topological) location, the network conditions at the time, server load, etc.  However, that is just the hostname.  Once you ask for a web page, the "localization", e.g. the banner add for the pizza place down the street, is determined by the IP address of the machine that does the HTTP GET.  (At least for all localizations I have seen.)

Hence my comment that Akamai, and as far as I know most large websites, do not use the location of the name server for things like language choice. 

-- 
TTFN,
patrick