Customer AS

Randy Bush writes:

>> just what i always wanted, two connections to a broken provider. you
>> must be kidding.
> I guess this brings to mind the question: Why would you want *any*
> connections to a broken provider?

all providers break.

Most providers also pretend this doesn't happen. One thing I'm
increasingly impatient with is the following:

You call up operations center for provider X.

"Why is it that I'm having trouble contacting node N?"
"Well, the router in Foostown is down."
"What do you mean, *THE* router in Foostown? You only have ONE!?"

When I've worked on building internal networks for banks, we've
typically done "no single point of failure" designs. Two links
everywhere over diverse paths, two routers carrying the traffic from
every point, etc. This means that no single router or piece of
equipment or cable dying can bring you down. If you architect things
right, it doesn't even cost that much more -- routers end up handling
disjoint subsets and the like. The real expense is multiple lines, but
that can often be used as a way to upgrade your bandwidth.

We can't pretend to be able to replace the phone networks until we can
achieve similar reliability. Phone networks typically are spec'ed to
two minutes a year downtime. We have a giant advantage in so far as
our protocols allow us to do failover far more gracefully. People
should start taking advantage of that.

The fact that things aren't nearly that reliable yet is the reason
people are forced to multihome their customer networks.