Potential role(s) of the "sub-IP" technologies in Future large-scale Network Architecture
by Dimitri Papadimitriou, Alcatel Bell, B-2018 Antwerpen, Belgium

Abstract: 

Traditional communication networks were segmented into circuit and packet
technologies (the latter with two variants: switched / connection-oriented and
connectionless forwarding). The role of IP / network layer was from its
inception developed to provide for a scalable, robust, application generic and
transparent overlay to interconnect networks of different technologies
(any-IP-any). Nowadays, sub-IP technologies are limited to traffic aggregation
in metro-/core-networks that interconnects IP edge/core devices. Over time,
the role of "sub-IP" technologies has evolved until reaching several
limitations.

 * First, pure circuit-oriented technologies imply fixed amount of resource
   reservation with hard-state cross-connection providing for point-to-point
   or point-to-multipoint connectivity between terminating network
   nodes. Hence, the "sub-IP" technologies have evolved such as to provide
   more flexibility from the inherent limitation of implying a strict
   relationship between connection and connectivity. Sub-IP technologies have
   evolved towards packetisation in the forwarding plane itself. Example: the
   control-driven MPLS paradigm has been extended in the optical domain as
   Optical Label Swapping (OLS), and the data-driven ATM paradigm has been
   extended in the optical domain as Optical Burst Switching (OBS).

 * Secondly, starting from manual provisioning even the unified control theory
   developed in the early 2000's (implemented e.g. via Generalized MPLS) has
   shown several limitations. Clearly the idea of providing a "control plane"
   is the preliminary form of a "packetisation" of the sub-IP
   environment. Indeed, the driver was to achieve a sub-IP environment being
   responsive and adaptive to the data traffic demands by mimicking behavior
   of IP/MPLS devices.  Nevertheless, over time several limitations have
   become more apparent: these range from the complexity of the protocol (due
   to the side effect of the generalization of the MPLS-TE protocol suite),
   the impossibility for pure IP device to setup data path without support of
   "switching capable interface" and more importantly the limitation induced
   by the MPLS control-driven paradigm itself.

 * Finally, Ethernet technology has progressively invade metro-/core-networks
   space and is nowadays positioned as the intra-domain interconnection
   technology of choice for aggregation networks by better adapting transport
   to Ethernet - as MPLS is adapted to IP. Nevertheless, moving Ethernet
   "networking" properties (tightly linked to LAN, and campus networks) toward
   metro-/core- networks will definitely transform the intrinsic nature of
   Ethernet and its control paradigm. Hence, we assist in an increasing
   interest in combining some form of constraint-based routing with an
   Ethernet forwarding plane not relying on MAC flooding/learning.

This presentation analyzes the tradeoffs in positioning different sub-IP
technologies for different large-scale future architectural scenarios as well
as the technical challenges associated to their foreseeable and combinable
roles. It will underline, the fundamental elements these technologies will
have to consider to offer a significant cost/gain and cost/perf ratio in
interconnecting IP devices (and not be limited to framing between IP
forwarding device's interfaces). This presentation will conclude by providing
modeling elements of the behavior of the end-to-end IP layer on top of a
revised set of sub-IP technologies.