[312] in Hesiod

home help back first fref pref prev next nref lref last post

Re: Info please

daemon@ATHENA.MIT.EDU (Amit Bhatiani)
Fri Jul 19 21:44:02 1996

Date: Fri, 19 Jul 1996 21:35:32 -0400
From: Amit Bhatiani <bhatiani_amit@jpmorgan.com>
Reply-To: amit@jpmorgan.com
To: "Theodore Y. Ts'o" <tytso@MIT.EDU>
Cc: asafier@csc.com, hesiod@MIT.EDU

Theodore Y. Ts'o wrote:

> If you want real high availability, then you should have automatic
> failover of your rounter of your entire network path to the rest of the
> outside world.  This then isn't a DNS issue, but a matter of having your
> router be multi-homed and having multiple points of connection to the
> Internet.
> 
umm..not really. The issue here is not the router, but the primary
server. Having the router be multi-homed and having multiple points of
connection to the internet still does not solve the problem of having a
fault-tolerant application server that transparently (to the
application's clients) switches over to the secondary application
server. This has nothing to do with the router or the internet
connection.

Also, having a multi-homed router and multiple points of connections
answers the question in a very non-general manner. What if this was an
application that just lived on a LAN and did not have any interaction
with the outside world whatsoever?

> That's because the DNS assumes that the network will provide whatever
> fault-tolerance you're looking for.  And providing high-quality fault
> tolernate in the IP protocol is a well understood problem ---- the
> solutions may be rather expensive, but they're well-understood.
> 

While I agree that DNS assumes fault-tolerance, it does not specifically
assume it to be in the network, your resolv.conf does allow for failover
within a set of primary servers for your zone and also you could provide
for the same mechanism for fault tolerance (as described previously for
high availability) for your DNS servers, at which point you would be
treating BIND as the application that needs to be fortified.

This is more of a matter of providing fault-tolerance at the application
layer than at the IP layer.

--amit


home help back first fref pref prev next nref lref last post