0 cellpadding=0 cellspacing=0 width=100%>
Yahoo! is not affiliated with the authors of this page or responsible for its content.
Terabit Burst Switching
Appears in
Journal of High Speed Networks, 1999.
Terabit Burst Switching
Jonathan S. Turner
Computer Science Department
Washington University
St. Louis, MO 63130-4899
jst@cs.wustl.edu
Abstract
Demand for network bandwidth is growing at unprecedented rates, placing growing demands
on switching and transmission technologies. Wavelength division multiplexing will soon make
it possible to combine hundreds of gigabit channels on a single ber. This paper presents an
architecture for
Burst Switching Systems designed to switch data among WDM links, treating each
link as a shared resource rather than just a collection of independent channels. The proposed network
architecture separates burst level data and control, allowing major simplications in the data path in
order to facilitate all-optical implementations. To handle short data bursts efciently, the burst level
control mechanisms in burst switching systems must keep track of future resource availability when
assigning arriving data bursts to channels or storage locations. The resulting
Lookahead Resource
Management problems raise new issues and require the invention of completely new types of high
speed control mechanisms. This paper introduces these problems and describes approaches to
burst level resource management that attempt to strike an appropriate balance between high speed
operation and efciency of resource usage.
1
Introduction
By some estimates, bandwidth usage in the Internet is doubling every six to twelve months [4].
For the rst time, data network capacities are surpassing voice network capacities and the growing
demand for network bandwidth is expected to continue well into the next century. Current networks
use only a small fraction of the available bandwidth of ber optic transmission links. The emergence
of WDM technology is now unlocking more of the available bandwidth, leading to lower costs which
can be expected to further fuel the demand for bandwidth.
1 We now face the near-term prospect of single bers capable of carrying hundreds of gigabits
per second of data. Single optical bers have the potential for carrying as much as 10 terabits per
second. This leads to a serious mismatch with current switching technologies which are capable of
switching at rates of only 1-10 Gb/s. While emerging ATM switches and IP routers can be used
to switch data using the individual channels within a WDM link (the channels typically operate at
2.4 Gb/s or 10 Gb/s), this approach implies that tens or hundreds of switch interfaces must be used
to terminate a single link with a large number of channels. Moreover, there can be a signicant loss
of statistical multiplexing efciency when the parallel channels are used simply as a collection of
independent links, rather than as a shared resource.
Proponents of optical switching have long advocated new approaches to switching using optical
technology in place of electronics in switching systems [2, 6, 9]. Unfortunately, the limitations of
optical component technology [7, 8, 12] have largely limited optical switching to facility manage-
ment applications. While there have been attempts to demonstrate the use of optical switching in
directly handling end-to-end user data channels, these experiments have been disappointing. Indeed
they primarily serve to show how crude optical components remain and have done little to stimulate
any serious move toward optical switching.
This paper proposes an approach to high performance switching that can more effectively exploit
the capabilities of ber optic transmission systems and could facilitate a transition to switching
systems in which optical technology plays a more central role.
Burst Switching is designed to make
best use of optical and electronic technologies. It uses electronics to provide dynamic control of
system resources, assigning individual user data bursts to channels of a WDM link. The control
mechanisms are designed to efciently handle data bursts as short as a kilobyte or as long as many
megabytes. Burst switching is designed to facilitate switching of the user data channels entirely
in the optical domain. While current optical components remain too crude for this to be practical,
anticipated improvements in integrated optics should ultimately make optical-domain switching
feasible and economically sensible. In the meantime, the data paths of burst switching systems can
be implemented in electronics with a substantial advantage in cost, relative to conventional ATM
or IP switches.
2
Network Architecture
Figure 1 shows the basic concept for a terabit burst network. The transmission links in the system
carry multiple
channels, any one of which can be dynamically assigned to a user data burst. The
channels can be implemented in any of several ways. Wavelength Division Multiplexing (WDM)
shows the greatest potential at this time, but Optical Time Division Multiplexing (OTDM) [10] is also
possible. Whichever multiplexing technique is used, one channel on each link is designated a
control
channel, and is used to control dynamic assignment of the remaining channels to user data bursts.
In addition to WDM and OTDM, there is a third alternative, which is not to multiplex channels on a
single ber, but to provide a separate ber for each channel in a multi-ber cable. Integrated optical
device arrays using optical ribbon cables that are now becoming commercially available [11], make
this option particularly attractive for LAN applications where cost is a predominant concern. In
fact, even in a wide area network using WDM or OTDM on inter-switch trunks, access links may
be most cost-effectively implemented using this approach.
2 KHDGHU FHOO
FRQWURO FKDQQHO
GDWD EXUVW
Figure 1: Terabit Burst Network Concept
The format of data sent on the data channels is not constrained by the burst switching system.
Data bursts may be IP packets, a stream of ATM cells, frame relay packets or raw bit streams
from remote data sensors. However, since the burst switching system must be able to interpret the
information on the control channel, a standard format is required there. In this paper, ATM cells are
assumed for the control channel. While other choices are possible, the use of ATM cells makes it
possible to use existing ATM switch components within the control subsystem of the burst switch,
reducing the number of unique components required.
When an end system has a burst of data to send, an idle channel on the access link is selected,
and the data burst is sent on that idle channel. Shortly before the burst transmission begins, a
Burst
Header Cell (BHC) is sent on the control channel, specifying the channel on which the burst is
being transmitted and the destination of the burst. A burst switch, on receiving the BHC, selects
an outgoing link leading toward the desired destination with an idle channel available, and then
establishes a path between the specied channel on the access link and the channel selected to carry
the burst. It also forwards the BHC on the control channel of the selected link, after modifying
the cell to specify the channel on which the burst is being forwarded. This process is repeated at
every switch along the path to the destination. The BHC also includes a length eld specifying the
amount of data in the burst. This is used to release the path at the end of the burst. If, when a burst
arrives, there is no channel available to accommodate the burst, the burst can be stored in a buffer.
There are two alternative ways to route bursts. In the
Datagram mode, BHCs include the
network address of the destination terminal, and each switch selects an outgoing link dynamically
from among the set of links to that destination address. This requires that the switch consult a
routing database at the start of each burst, to enable it to make the appropriate routing decisions.
In the
Virtual Circuit mode, burst transmissions must be preceded by an end-to-end route selection
process, similar to an ATM virtual circuit establishment. During route selection, the forwarding
information associated with a given end-to-end session is stored in the switches along the path.
3 BHCs include a Virtual Circuit Identier (VCI), which the switches use in much the same way that
ATM switches use an ATM VCI. Note that while the route selection xes the sequence of links used
by bursts in a given end-to-end session, it does not dictate the individual channels used by bursts.
Indeed, channels are assigned only while bursts are being sent.
To handle short data bursts efciently, burst switching systems must maintain tight control over
the timing relationships between BHCs and their corresponding data bursts. Uncertainty in the
precise timing of the beginning and ending of bursts leads to inefciencies, since burst switching
systems must allow for these uncertainties when setting up and tearing down paths. For efcient
operation, timing uncertainties should be no more than about 10% of the average burst duration.
For efcient handling of bursts with average lengths of 1 KB or less on 2.4 Gb/s channels, we need
to keep the end-to-end timing uncertainty in a burst network to no more than 333 ns.
3
Key Performance Issues
There are two key perf