Is caching necessary?

[Kevin Darcy]

Brad Knowles wrote:

>At 11:45 PM -0400 2005-04-29, Kevin Darcy wrote:
>
>  
>
>> The only legitimate reason for forwarding to a central cache, when one
>> has the option available to query Internet nameservers directly, is if
>> the constellation of local network topology, query patterns, TTL values,
>> etc. happen to all align so as to make your average and/or worst-case
>> query latency better that way.
>>    
>>
>
>	No, there's another legitimate reason.  When you have a business 
>need to guarantee that all internal servers see the same answers for 
>a given question (modulo TTLs), but you can't point all those servers 
>directly at a single caching/recursive server.
>
>	For example, at AOL I had a problem where mail would come in from 
>a given sender to a given recipient via one particular server, and 
>everything would work fine.  However, mail from the same sender to 
>the same recipient via a different server would find different 
>information in the DNS (perhaps the delegation was broken, or there 
>was a lame server, or somesuch), and we would end up rejecting the 
>message.
>
>	Try doing that with ten million users, handling tens of millions 
>of messages per day, and even if this kind of thing only has a 
>one-in-a-thousand chance of happening, that means you have a very 
>large number of extremely irate customers who are all looking to use 
>explosive devices in new and creative ways.
>
>
>	Performance is one valid reason where forwarding may be used, but 
>there are others.
>
Perhaps I should have qualified that as "The only legitimate technical 
reason ...". What you describe sounds more like a user-expectation 
problem, not really a techinical issue _per_se_. It is a well-known 
technical fact that DNS data propagates unevenly to different parts of 
the Internet infrastructure at different times.

- Kevin