BCP for multisite multihoming

Steve Powell spowell at gblx.net
Mon May 21 23:46:12 CEST 2007


Greetings.  I might have missed this as the response
to the entire multihoming question has generated a
lot of interest.  I am just throwing knee jerk ideas out.
If you are trying to do something with multihoming that
requires users to make a change, you have to wait for a
billion hosts to update.  Seems like making the routers
do something unorthodox is kind of hard.  You can sort of
do load balancing by multi addressing a host and depend
on a DNS rotary.  If you have 2 providers and 2 address
ranges, you can in theory load balance.

The problem is that the scope of DNS is pretty universal.
You can't tell a customer of ISP A, which is also the provider
for our duel homed entitiy to NOT use ISP B, if the coin
on the rotary flipped and said use B.  If there was some
kind of selective scoping, a customer could  reply to a query
not just on randomness or round robin, but based on the source
address of the query.

This would call for a lot of changes, and despite my limited
DNS foo, can think of a dozen reasons why this idea might
be considered stupid.  But if you could bypass those hurdles
and have DNS servers make decisions based on routing information,
you could preserve a small routing table.  It seems to be a
lot of work.  But if keeping state low in the routers is very
important, and it cannot be handled by the host, then the devices
that handle DNS are the only ones left to have the increased
entropy offloaded to.  At the very least there are probably fewer
DNS servers then routers in the world(just a perception).  But
would DNS admins enjoy the extra responsibility of traffic
engineering?

Then again they say that transmision and routing is converging.
Maybe the future will be one giant monster of a box that does
my transmision, routing, DNS, and runs microsoft office ;)

stevep



More information about the ipv6-ops mailing list