[rrg] 2-Phased Mapping - alternative critique

[Robin Whittle <[e-mail]>]

The RRG Report ID currently contains a simple critique of 2-Phased
Mapping by the designer - Wei Zhang.

Here is an alternative critique - 455 words.

  - Robin



Documents
---------

  Proposal with slides:

     http://www.ietf.org/mail-archive/web/rrg/current/msg05536.html

  Discussion:

     http://www.ietf.org/mail-archive/web/rrg/current/msg05713.html

  RRG Report ID, with critique:

     http://tools.ietf.org/html/draft-irtf-rrg-recommendation-04#section-9

     http://tools.ietf.org/html/draft-irtf-rrg-recommendation-04#section-9.2




Critique
--------

2-Phased Mapping (2PM) proposes a novel arrangement for reducing the
load on a global mapping system for a Core-Edge Separation architecture
- in which ITRs look up the mapping of a packet's "edge" destination
address, as part of determining which ETR to tunnel the packet to.

2PM is not a complete proposal, so it cannot be considered as a solution
to the routing scaling problem.

There appears to be an assumption inherent in 2PM that all networks
which use scalable "edge" prefixes have Autonomous System numbers.
Since scalable routing solutions may need to support ten million or so
non-mobile end-user networks, each with one or more scalable "edge"
prefixes - and potentially ten billion mobile devices - it is reasonable
to assume that most of these end-user networks will not have ASNs.

2PM aims to reduce the total number of prefixes in the global mapping
system by each end-user network (which is assumed to have an ASN)
placing not its scalable "edge" prefixes in the system, but one or more
covering prefixes for these "edge" prefixes, together with its ASN.
This is also intended to reduce the update rate for the global mapping
system.

ITRs would query the global mapping system with a packet's destination
address, and receive an ASN number.  ITRs would have previously
discovered, via a new BGP extension, the closest ETR for each ASN.  The
ITR would then tunnel the packets to that ETR.

There are a number of problems with this.  Firstly, the whole idea of
the "portability" aspect of scalable routing is that an end-user network
could use its scalable "edge" space via any ETR in the world.  So the
assumption that all its prefixes would be mapped to one, or several,
ETRs operated by the end-user is incorrect.

Secondly, the use of the DFZ's control plane to propagate to all ITRs
the best paths to one or more of an ASN's ETRs is at odds with a primary
aim of scalable routing: to reduce the load on the DFZ's control plane.

Thirdly, by tunneling to the closest ETR, it is not clear how the ITR
would discover whether or not the destination host was reachable via
this ETR - or if it did, how the ITR would select an alternative ETR.

Furthermore, CES architectures generally do not require ITR and ETR
functions to exist on busy DFZ routers.

2-Phased Mapping appears to be an unsuitable alteration for CES mapping
systems.  Nonetheless, alternative arrangements for mapping systems are
an important contribution to the design process, by exploring approaches
which turn out not to be useful, and perhaps also to suggest new
approaches which, with further development, do have significant advantages.