Cell Relay Archive

["Albert Manfredi" <albert.e.manfredi@boeing.com>]

"Aron Wahl" <awahl@yahoo.com> wrote:

> Euuh according to what I know MPOA and RFC1483 are 2 things
completely
> different!!!!(I am supposed to know it cause I have implemented
> RFC1483). RFC1483 is just a list of headers you have to add to
frames
> of various protocols when sending over ATM. LANE uses RFC1483 as
> encapsulation method to simulated a LAN over ATM. And MPOA uses
LANE.
> And as far as I know MPOA also is supposed to detect IP flows when
> reaching a certain number of packets. MPOA will create, in that
case,
> a shortcut virtual circuit directly to the destination MPOA
router. It
> also separates the routing from the forwarding no???

MPOA is basically LANE with shortcuts available. Where LANE is the
simple application of ATM to the role of an IEEE 802 LAN, and MPOA
adds a routing function to LANE, to allow ATM to be more than just a
little island.

A similar approach is RFC 2225 + RFC 2332, i.e classical IP over ATM
+ next hop resolution protocol. The difference between MPOA and
these two RFCs is that the CLIP + NHRP do not try to emulate the
IEEE 802 link layer. These RFCs instead run IP over ATM directly.

But MPLS is something else. It builds on the idea of using "labels,"
sort of like VPI/VCI labels, to allow IP routers to send packets
down pre-defined tunnels without having to look at routing tables.
So yes, it's pretty straightforward to implement something like MPLS
over an ATM network, perhaps using ATM PVCs or SPVCs, but MPLS does
not require ATM.

I always considered MPOA or CLIP+NHRP to be ways of integrating IP
and ATM very closely. Actually making use of ATM features to benefit
IP. Whereas I always considered MPLS to be the conceptually much
simpler approach of building static tunnels. Using ATM this way is
probably a lot more pragmatic and widespread, but also less
interesting.

There are now new efforts into something called generalized MPLS, or
GMPLS. This woulkd add a UNI interface to MPLS, to actually permit
the end user to signal to the network core, to set up MPLS paths in
real time. This would get MPLS a lot closer conceptually to MPOA or
CLIP + NHRP, I think. The question will be, if there were scaling
problems in MPOA and NHRP, why would these same problems not rear
their ugly heads again, with GMPLS?

Bert
albert.e.manfredi@boeing.com