Re: UDP ports and the ITU Liason

["W. Mark Townsley" <[e-mail]>]

Eric Rosen wrote:

> "The use  of pseudowires  over UDP/IP  is only applicable  where there  is a
> technical constraint preventing the use of L2TPv3 or GRE / MPLS." 
> 
> The technical  constraint being  what, the fact  that a product  already has
> support for UDP and UDP only? 

Isn't that roughly the same reason we give in 
draft-ietf-mpls-in-ip-or-gre-08.txt for MPLS over GRE vs. MPLS over IP?

      - Implementation considerations may dictate the use of MPLS-in-GRE.
        For example, some hardware device might only be able to handle
        GRE encapsulations in its fastpath.

> 
> It seems to me  that this WG has not given any  serious consideration to the
> use of UDP in this manner, and hence should not be making any recommendation
> to the ITU-T with regard to its use. 

The most practical argument one might have for using UDP when tunneling over IP 
is to be more friendly to NAT. For example, draft-ietf-ipsec-udp-encaps-09.txt 
does precisely this. L2TPv3 also has an optional mode that inserts UDP for NAT 
friendliness. Now, being NAT friendly generally means one UDP port mapping per 
host behind the NAT. Overloading the UDP port number with the PW multiplexer 
serves to tax NAT tables to their extreme, ultimately reducing the number of 
sessions across the set of multiple applications which use UDP behind a given 
NAT. Yes, I know voice uses UDP. That's part of the problem. What if *all* 
applications used UDP ports for session multiplexing?

> 
> The proposed liaison statement appears  to be a conditional endorsement of a
> proposed ITU-T standard, but without  any mechanism to declare that standard
> to also be an IETF standard.  Frankly, I don't understand this procedure.  
> 
> The liaison  is further confusing in  that it declares certain  things to be
> the consensus of the  WG, and then goes on to state  the various reasons why
> the WG hasn't achieved a consensus. 
> 
 > Perhaps a better liaison statement would  be "the PWE3 WG has not considered
 > the use of UDP as the tunneling  mechanism for PWE3, and at the present time
 > is not planning a standards track document on that topic."

But the record of emails, at least one draft vying for standards track on the 
topic, and mention of UDP specifically in our charter would all suggest that we 
have considered UDP.

- Mark