[AFS3-std] DNS SRV Resource Records for AFS

Jeffrey Hutzelman jhutz@cmu.edu
Mon, 05 Oct 2009 21:15:54 -0400


--On Monday, October 05, 2009 04:53:51 PM -0700 "Buhrmaster, Gary" 
<gtb@slac.stanford.edu> wrote:

>> Ultimately, there's some interesting things that can be done with load
>> balancing if TTLs are obeyed, thus the question.
>
> Note that in practice a number of ISP's will "adjust" the TTL
> on their caching DNS servers, regardless of what the original
> TTL is(*).

True.  And in fact, they may do so in both directions, not only discarding 
data whose TTL has not expired (which is perfectly legal) but also caching 
data well beyond its TTL.  Such behavior is noncompliant, but some 
providers don't care; this is one reason I prefer not to use a caching 
resolver not under my control.  Nonetheless, part of the point here is that 
behavior such as this _breaks_ DNS-based load-balancing, sometimes quite 
badly.

> And when there are multiple levels of DNS caching,
> the practical TTL may end up being longer.

That shouldn't happen, but yes, it sometimes does.

Also, TTL's of zero are not always safe to use, because some resolvers will 
throw them away too quickly.


-- Jeff