Re: Host-based may be the way to go, but network controls areneccessary

[Iljitsch van Beijnum <iljitsch@muada.com>]

On Thu, 21 Nov 2002, Kurt Erik Lindqvist wrote:

> > Future: when it becomes possible for a host to jump from one set of
> > addresses to another (as in SCTP) it certainly becomes useful to do
> > this. The crude way: drop packets, the host will get the hint. The nice
> > way could be an ICMP message.

> That means that the router would have to keep even more state than
> today so it would know when to send an update to the host.

I'm not sure what you mean by "know when to send an update". When the
router (or other box designated for this function) sees too much traffic
over one link it just starts sending back ICMPs to redirect some of it.

> State is among the most costly thing for network operators, you want to
> avoid it. To make this scale you would have to send periodic updates
> and the host would know only when the next update comes.

We're not talking about network operators here, but about enterprises.
If they want this, they'll have to pay for it. Just like traffic
engineering in BGP today costs money in the form of man hours.

> So with packet-loss or the above, there will first be a service impact

Unreliable datagram service. Sometimes you lose one.

> Why do I get that MPLS feeling here...

I don't know. When the only tool you have is a hammer, everything looks
like a nail?

Iljitsch