The-Public-Internet-The-Second-Hierarchy

The Internet, as you have seen, is built largely on basic telecommunications infrastructure. All,
or almost all, signals pass over the same basic physical infrastructure of last-mile twisted-pair
copper and metro and long-haul fiber as do voice calls, but once onto long-haul fiber or, in some cases, metro fiber, ordinary circuit telephone switches do not determine the direction of
the data traffic any longer. Rather, the switch sends the data traffic to a network access point
(NAP) in what is usually a local telephone call, and from there circuit-switching procedures of
the sort utilized in traditional telephone networks are no longer invoked. Instead, a core router
assumes the function of directing data traffic through the network.
Core routers are specialized switches that switch individual packets rather than complete
streams, and, in the case of best-effort traffic without stringent QoS requirements, the individual
packets may take different routes through the network to avoid congestion.
Routers themselves may be yoked with ATM switches or may even incorporate ATM functionality,
and in such cases the ATM protocol may be used to encapsulate IP traffic. This results
in some loss of efficiency but, for reasons that go beyond the scope of this book, aids in shaping
traffic and managing bandwidth. I see ATM gradually losing ground to MPLS, which has similar
capabilities but offers greater efficiency and much more flexibility when it comes to service
creation. The tendency today in the core routers used at major Internet hubs is to combine the
functionality of an MPLS switch with an Internet router in a single box.
NAPs and their associated routers are generally owned and operated by long-distance
carriers, though a number of independents exist such as Equinix, Focal Communications
Corporation, and TelX that maintain and operate NAPs of their own. Most such facilities aim
to serve independent ISPs, but in some cases local competitive access providers also will be
accommodated.
If wireless broadband operators want to confine themselves to offering nothing more
than local high-speed access, none of this matters terribly much, but if they want to offer or
support conferencing (particularly videoconferencing), streaming media services, Web hosting
services, IP storage, real-time interactive applications such as multiplayer gaming, and
pure end-to-end IP telephony, then the nature of the Internet connection becomes crucially
important.
The Internet, as I have indicated previously, was conceived as a best-effort delivery
network where the predictability of a connection was sacrificed in the interest of overall
robustness and redundancy. The Transmission Control Protocol/Internet Protocol (TCP/IP)
suite was never designed to support low-latency, constant bit rate transmissions, and although
some of the older ancillary protocols such as UDP provided some support for delay-sensitive
applications, the Internet never matched the ATM networks set up by the large telcos in terms
of QoS.
Because of this basic deficiency, the major router manufacturers (Cisco Systems and
Juniper Networks) have strongly supported standards (principally MPLS, RSVP, and DiffServ)
that would enable routers to vie with ATM switches in supporting full QoS across a wide range
of applications. Indeed, some would contend that current MPLS router/switches actually do a
better job of supporting differentiated services than do legacy ATM switches.
Today most of the large carrier-class IP routers on the market are MPLS enabled and can
support QoS for multimedia and real-time interactive applications, but that does not mean
that a high-speed, high-fidelity multimedia transmission sent out over the public network is
necessarily going to arrive intact at its destination. QoS enforced through the implementation
of MPLS will result in more efficient use of bandwidth than would be the case carrying IP over
ATM, but it still involves reservation of bandwidth, which means that bandwidth cannot be
made available to anyone else until the transmission is finished. Time-sensitive transmissions
simply make greater demands on capacity, and capacity always comes at a price particularly in
the metro and the last mile. Long-distance carriers can choose not to meet those demands. MPLS functionality can be partially or completely disabled in a router, and since ISPs and
long-distance service providers are loathe to tax capacity to support QoS, delay-sensitive traffic
is apt to encounter such a disabled router in its journey across the Internet to its final destination.
If that occurs, it does not matter how stringent QoS provisions were in the path already
traversed. The transmission essentially returns to best effort.
Some ISPs and long-distance carriers are willing to negotiate service-level contracts with
local broadband access providers that will provide end-to-end QoS over an MPLS backbone. If
broadband wireless operators want to launch transmissions outside their own network that
require more than best effort, then such agreements are essential.