Re: [Pppext] [Int-area] PPP-to-ethernet

[James Carlson <james.d.carlson at sun.com>]

Derick Winkworth writes:
> Yeah, that last e-mail was not well thought out.  Lets ignore that and refer to the first one I sent.
> 
> The original thought was just for router-to-router traffic, and the idea was, in the case of dissimilar media-types, to extend the functionality of local-switching/tcc to allow the remote PPP peer to form routing adjacencies to more than one router on an ethernet segment.  

I'm not sure why anything special is needed for that.  It's
commonplace to run routing protocols over PPP links and not too
unusual to use proxy-ARP'd links.

But to what benefit?  A PPP link is a single interface.  You don't get
any routing-related advantages by pretending that it does anything
other than connecting two distinct nodes together, and thus has
exactly two peers, do you?

My guess is that you're trying to cover for a lack of routing prowess
in the box that terminates the PPP link itself.  Adding complexity in
the form of tunneled links doesn't sound to me like a good way to fix
that problem.

> The reason, supposedly, for local-switching/tcc in the first place is to allow for L2VPN service within the same device in the same POP.  Allowing this for dissimilar media types of course means that sites which may be an aggregation point for hundreds of PVCs or PPP links can be connected via ethernet rather than many circuits/channels/etc.  There is also the case that some sites simply may only be reachable with a particular media.

I don't understand.  This is done *today* using L2TP.  It's even done
using the non-standard PPPoE.  What new mechanism is required?

> At this point, the switching of dissimilar media types is done strictly on a "point-to-point" basis and in our own testing redundancy and fail over at the head-end ethernet connected site does not work well unless you are doing static routing and pointing to a VRRP or HSRP address.  If you want dynamic routing, then it doesn't work as gracefully (or at all).  

Really?  That sounds like an equipment limitation or bug rather than a
statement about the protocol itself.

RIP-2, OSPF, IS-IS, and BGP all work fine for me over PPP links.  HSRP
and VRRP are inapplicable for point-to-point links, as they're not
broadcast-type links, but why would you need those special "dumb host"
hacks anyway?

I'm confused about what problem is really being solved here.

> Here we thought, it would be great for scalability and redundancy if we could somehow allow the remote PPP device to peer, as I said, to multiple routers on the ethernet segment in a datacenter or across datacenters that are connected via ethernet.  The address field in the PPP packet would always represent a router on the ethernet segment (either the router it is going to, or the router it came from).  On a given ethernet segment, multiple remote PPP devices would have virtual-mac addresses on the ethernet segment which the PE device is listening for and has mappings for to the remote PPP devices.

I don't agree with your stated rationale for doing this (it doesn't
really help at all with routing), but the solution you're describing
is already available.

> So the remote devices would need to support this functionality, and the PE devices in the CO would need to do the actual translating.
> 
> I am working on a presentation with diagrams at the moment.  I'll send a link.  I hope though that the above clarifies this somewhat.

Somewhat.  But you still haven't explained why the existing tunneling
protocols are insufficient for the task.  I don't think we should
create any new protocols until we understand why the existing ones
don't do the job they were designed to do.

-- 
James Carlson, Solaris Networking              <james.d.carlson at sun.com>
Sun Microsystems / 35 Network Drive        71.232W   Vox +1 781 442 2084
MS UBUR02-212 / Burlington MA 01803-2757   42.496N   Fax +1 781 442 1677
_______________________________________________
Pppext mailing list
Pppext at ietf.org
http://www.ietf.org/mailman/listinfo/pppext