Re: Caching Architectural Considerations

From: James R Grinter <jrg@dont-contact.us>
Date: Fri, 1 Nov 1996 00:48:43 +0000

On Thu 31 Oct, 1996, Oskar Pearson <oskar@is.co.za> wrote:
>won't take 8 hours for everything else to know about it. You can probably
>set it to something like 600 seconds. This way the worst thing that could
>happen is that some people are down for 10 minutes...

nice ideas, but I've been experimenting today with the behaviour of
Netscape (v3.0b1, under Unix, only, so far) and DNS lookups, and it
isn't good.

It would seem that it uses only a gethostbyname() call for the cache
address - using an address with two A records leaves Netscape using
only the first one in the (random order) response that it got (verified
by using tcpdump to look at the nameserver responses). If it *is* using
gethostbyname then it won't get the ttl information either.

Further it caches that response and doesn't look it up again. When
changing the Address record for the name, a day old Netscape didn't
notice and failed to get a response from the (now old) address.

Using Netscape's proxy configuration file gives some extra facilities:
by listing the combined address, and then each of the individual hosts
as fall backs I could get the behaviour I wanted, but possibly with two
timeout delays in the worst case, if Netscape didn't quickly realise
which address had already failed.

Anyone looked at other platforms, other versions, or other browsers?
I'm also curious as to what Internet Explorer does (the other most
commonly used browser, of course). Anyone want to help produce a
summary of current practice and produce a best practice document. (Has
anyone done this already, recently? I do remember it being talked
about)

James.
Received on Thu Oct 31 1996 - 16:49:15 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:33:25 MST