cache and ttl, resolvers and clients

[Barry Margolin]

In article <378CC86D.7D69B88 at mediaone.net>,
Robert Bowles  <rbowles at mediaone.net> wrote:
>Are there many pitfalls with various dns clients/resolvers honoring (or not
>honoring) ttl?  What if the ttl values are on the order of 1-2 minutes?

Most resolvers don't return TTL information, so clients don't know what it
is they're supposed to honor.  For instance, take a look at the
gethostbyname() API -- there's no TTL in it.  If you assume that TTLs are
only honored by caching DNS servers you'll be pretty close to the mark.

Many applications have their own caches, usually on the order of 5 minutes.
This allows something like Netscape to avoid doing repeated lookups for all
the URLs that are on a page and point back to the same server.

>Its easy enough to verify (I'm specifically interested in Solaris' nscd
>honoring small ttls without having to turn caching off altoghether).

nsdc's TTL can be configured in nscd.conf.  Check the man page.

>I have seen issues with the Java runtime maintaining its own DNS resolution
>cache and it did not appear to honor ttl at all.  (And in fact would only hit
>the Solaris resolver when the runtime was restarted.)

I think early versions of Netscape didn't have a timeout, either.  The
programmers apparently didn't imagine people leaving the same browser
session running for days or weeks (maybe they were too used to MS Windows,
so they couldn't imagine the computer staying up that long).

-- 
Barry Margolin, barmar at bbnplanet.com
GTE Internetworking, Powered by BBN, Burlington, MA
*** DON'T SEND TECHNICAL QUESTIONS DIRECTLY TO ME, post them to newsgroups.
Please DON'T copy followups to me -- I'll assume it wasn't posted to the group.