How Many Wavelenths Do We Really Need in an Optical Backbone Network?

Joe Bannister, Joe Touch, & Allan Willner, University of Southern California
Stephen Suryaputra, Nortel Bay Networks

Abstract

Coupling Internet protocol (IP) routers with wavelength-selective optical
cross- connects makes it possible to support existing Internet infrastructure
in a wavelength division multiplexing (WDM) optical network.  Because optical
wavelength routing is transparent to IP, one can achieve very high throughput
and low delay when packets are made to bypass the IP forwarding process by
being routed directly through the optical cross-connect.  We study the
performance of a specific instantiation of this approach, which we call
packet over wavelengths (POW).  We present the POW architecture in detail and
discuss its salient features.  Realistic simulations of the POW that use
actual packet traces in a well-known Internet backbone network reveal the
level of performance that can be expected from POW under various options.
Specifically, we evaluate the fraction of packets that are switched through
the cross-connect as a function of the number of WDM channels and the degree
of flow aggregation that can be achieved.

The deployment of wavelength division multiplexing links has begun, and it is
highly desirable to use these links to interconnect the routers that comprise
the global Internet.  We consider the POW network, in which packets are
forwarded by both IP routers and wavelength routers.  The goal of such an
architecture is to switch as much traffic as possible directly by means of
wavelength routers, because IP forwarding is relatively expensive by
comparison.  However, wavelength routing is limited by the fact that only a
few (eight to 64) WDM channels per link are supported by today=92s
technology.  Our intent is to study and characterize the expected performance
of POW.  To this end we examine different options for recognizing which
packets should be switched through a wavelength router and which packets
should be forwarded by an IP router.  We conduct simulations to determine the
level of WDM needed to carry a substantial fraction of packets in a switched
(rather than a routed) mode.

POW shares features with label switching.  Label switching is used when an IP
router includes a switching fabric that can be used to bypass IP forwarding.
Since switching speeds are much greater than forwarding speeds (estimated by
some to be 20 times greater for comparably priced hardware), one attempts to
place as large a fraction of packets as possible on the switched path,
leaving as small a fraction of packets as possible on the forwarded path.  To
accomplish this requires some above-average intelligence in the
switch-router.  The router must have software that recognizes that a flow of
packets can be passed through the switching fabric.  A signaling protocol
then assists in notifying switches that the recognized flow should be carried
over a switched path rather than a routed path.  Eventually a hop-by-hop
sequence of switches carries the flow of packets from one router to another.
There are currently a number of projects to evaluate and implement label
switching in WDM networks, so it is crucial to understand fully the
engineering tradeoffs that one encounters.

Our ns-based simulations of the NSF Very-High Bandwidth Network Service
(vBNS) backbone with real tcpdump traffic traces show that the use of current
WDM technology will not provide acceptable performance.  Achieving an
acceptable level of performance requires that POW be able to aggregate or
groom traffic so that flows can be switched en masse and still utilize WDM
channels effectively.  Our simulations show that aggregation by backbone
egress point can result in over 90% of all packets being directly switched
through the backbone.