Optical interconnection technology in switches, routers and optical cross connects (länk till annan webbplats)

Optical interconnection technology in switches, routers and optical cross connects (länk till annan webbplats)
5138454
6/23/03
10:44 AM
Page 20
Jonsson, M., “Optical interconnection technology in switches, routers and
optical cross connects,” SPIE Optical Networks Magazine, vol. 4, no. 4, pp.
20-34, July/Aug. 2003.
Optical
interconnection
technology in
switches,
routers, and
optical cross
connects
Magnus Jonsson
School of Information Science,
Computer and Electrical Engineering
Halmstad University
Halmstad, Sweden
ABSTRACT
The performance of data- and telecommunication equipment must keep abreast of the increasing network speed.
At the same time, it is necessary to deal with the internal
interconnection complexity, which typically grows by N2
or NlogN, where N is the number of ports. This requires
new interconnection technologies to be used internally in
the equipment. Optical interconnection technology is a
promising alternative and much work has already been
done. This paper reviews a number of optical and optoelectronic interconnection architectures, especially from
a data and telecommunication equipment point of view.
Three kinds of systems for adopting optical interconnection technology are discussed: (i) optical cross connects
(OXCs), (ii) switches and routers with some kind of burst
switching and (iii) switches and routers that redirect traffic on the packet or cell level. The interconnection technologies and architectures are discussed according to
their suitability for adoption in the three system types.
20
1 Introduction
N
ovel optical technologies offer possibilities for
new solutions for the increasing bandwidth
demands of data communication and telecommunication equipment. This paper uses a primarily
architectural approach to explore the possibilities of
using optical interconnections in such equipment. Some
selected groups of concepts are selected here in order to
give a reasonably broad view in the scope of a single
paper.
The interconnection architectures are evaluated
according to three system types with different switching
time requirements. These types are: (i) optical cross connects (OXCs), (ii) switches and routers with some kind of
burst switching and (iii) switches and routers that redirect
traffic on the packet or cell level.
OXCs have relatively slow timing requirements, i.e. in
the order of milliseconds or even tens of milliseconds. On
the other hand, it is valuable if the signals remain optical all
the way through the switch, including possible queuing
systems, i.e. there is optical transparency. In this way it is
easier to scale up the bit rate or change protocols.
With burst switching, packets with the same destination, like the same exterior gateway, can be grouped
together to reduce switching time requirements. This has
for instance been discussed for use in all-optical packet
switching [1,2]. The switching time requirements are in
the order of sub-microseconds.
For pure switching at the packet level, the switching
time requirements are in the order of nanoseconds. With
longer switching times, the overhead between the packets
will be too large. All-optical packet switching has been
proposed but much work remains to reach purely optical
routers [1]. However, some experiments have been done
and several architectures that are more or less purely
optical have been proposed [3-7].
The interconnection networks reviewed are intended
for use in, or as substitution of, the switch core in data or
telecommunication equipment. Other components needed
in addition to the switch core vary between different
equipment. As an example, a simple switch might be
implemented without much more components, while an IP
router needs components to manage the routing table. The
acceptable cost of an interconnection network can also vary
depending on whether it is to be used in a core router,
where many users share the cost, or in equipment near the
end-user.
Another driving application domain for optical
system-level interconnections, in addition to data and
Copyright 2003 Society of Photo-Optical Instrumentation Engineers and Kluwer Academic Publishers.
This paper was published in Optical Network Magazine and is made available as an electronic reprint with
permission of SPIE and KAP. One print or electronic copy may be made for personal use only. Systematic
or multiple reproduction, distribution to multiple locations via electronic or other means, duplication of any
material in this paper for a fee or for commercial purposes, or modification of the content of the paper are prohibited.
Optical Networks Magazine July/August 2003
5138454
6/23/03
10:44 AM
Page 21
telecommunication equipment, is parallel and distributed computing systems [8-11]. Important work in the
field of optical interconnections in computing and
communications started in the mid 1980s. Of the
early publications, a good paper on system aspects was
published by Goodman et al. in 1984 [12], while publications by Miller et al. can represent work on specific
components. Other examples of early work in the field
are [13-16].
The rest of the paper is organized as follows. Switch
fabrics are generally reviewed in Section 2, and Section 3
describes optical link technologies. Section 4 presents
fiber-optic interconnection networks. Integrated optical
interconnection systems and optical and optoelectronic
switch-fabrics are described in Sections 5 and 6, respectively. Sections 7 and 8 give a discussion and a summary,
respectively.
2 Switch Fabrics
An example of a switch or router is shown in
Figure 1, where the switch fabric transfers traffic from
input ports to output ports on the basis of decisions
made by some control logic. One of the simplest ways to
implement or simulate an electrical switch fabric is to
have a shared medium to which I/O interfaces, control
processors or the like are attached and over which the
traffic is time multiplexed (see Figure 2). A common way
of implementing a shared-medium network is to use the
bus topology, where only one node is allowed to send at
a time, but a shared-medium network can also be a ring
for example. The great advantage of a shared-medium
network is the easy implementation of broadcast, which
is useful in many situations. The disadvantage is that the
bandwidth does not scale with the number of nodes
(ports on a switch).
The crossbar is the most flexible switch fabric
and can be compared with a fully connected topology,
i.e. all possible pairs of input and output ports are
connected by point-to-point connections. The drawback, however, is the increase by N 2 in cost/complexity
of the switch, where N is the number of ports. Systems
with a single electrical true crossbar are therefore
limited to small or medium-sized systems. A 140 140 3.2 Gbit/s single-chip switch is commercially
available [17].
In multistage shuffle-exchange networks, the cost
function is reduced to N log2 N, but here log2 N stages
must be traversed to reach the desired output port. An
example of this kind of a network is the Omega network
(Figure 3), which provides exactly one path from any
input to any output. The different switch functions of the
2 2 switch that is used as the building block are shown
in Figure 4, where the two right most configurations are
used for broadcast. Switches larger than 2 2 can also be
used. Each stage of the switches in an Omega network is
preceded by a perfect-shuffle interconnection pattern. In
contrast to a crossbar network, which is a nonblocking
network, an Omega network is a blocking network. This
means that there may not always exist a path through the
network as a result of already existing paths that block the
way. Rearrangeable networks are another category of
multistage networks where it is always possible to find a
path through the network. However, if all paths are not
routed at the same time, it may be necessary to reroute
already existing paths.
3 Optical Link Technologies
This section discusses two main categories of optical
links, fiber-ribbon links and bit-parallel WDM links. Single-channel single-fiber solutions are not treated.
Figure 1: An example of switch architecture.
Figure 3: Eight-channel Omega network. One path through the
network is highlighted.
Figure 2: Bus-based switch architecture.
Optical Networks Magazine July/August 2003
Figure 4: Possible states of a 2 2 switch.
21
5138454
6/23/03
10:44 AM
Page 22
3.1 Fiber-ribbon links
A system component that reached the market some
years ago is the fiber-ribbon link. As an example of an
early commercial link, the Motorola OPTOBUS has
ten parallel fibers, each carrying data at a bit rate of
400 Mbit/s, which gives an aggregated bandwidth of
4 Gbit/s [18]. Bi-directional links are also possible, where
some fibers in the fiber-ribbon cables are dedicated for
each direction [19]. Further references to reports on fiberribbon links are found in [20].
In scaling up the bandwidth of a fiber-ribbon link
where a dedicated fiber carries the clock signal, the main
problem is channel-to-channel skew. The skew is mainly
the result of differences in propagation delay between
different fibers and variations of lasing delay time among
different laser diodes [21]. The 400 Mbit/s OPTOBUS
has a specified maximum skew of 200 ps, excluding the
fiber-ribbon cable for which 6 ps/m is assumed for standard ribbons [18]. Even though the distances are rather
short in the systems discussed in this paper, the scaling
to higher speeds calls for a discussion of techniques to
reduce the effect of the skew.
One technique is to actually reduce the skew, either
by using low skew ribbons or employing skew compensation. Fiber ribbons with a skew below 1 ps/m have
been developed [22], which essentially increases the
possible bandwidth distance product. All the fibers in
the same ribbon are sequentially cut to reduce the
variation in refractive index among the fibers. In the
fiber-ribbon link described in [23], a dedicated fiber
carries a clock signal used to clock data on 31 fibers.
The transmitter circuitry for each channel has a programmable clock skew adjustment to adjust the clock in
80-ps increments.
Another technique is to extract the clock signal from
the bit flow on each fiber instead of using a separate fiber
to carry the clock signal. The disadvantage is increased
hardware complexity when a clock recovery circuit and
a buffer circuit for each channel in the receiver are added.
A hybrid solution is to skip the separate clock channel
and encode clock information on the data channels while
still sending in bit-parallel mode, as reported in [24]. In
this case, a deskew unit relying on FIFO registers (First
In First Out) ensures that parallel data words that are output from the receiver are identical to those which were
sent. A possible 15-ns deskew was reported. A similar
system is reported in [25].
The techniques mentioned above introduce either
increased hardware complexity or a more sophisticated
fiber-ribbon manufacturing process. If the manufacturing process allows for adding more fibers in each ribbon, this may be a cheaper alternative. For example,
a fiber-ribbon link with 32 fibers, each with a bit rate
of 500 Mbit/s, was described in [23], and researchers
at NEC have developed a module in which 8 2 lasers
are coupled to two fiber-ribbons [26]. Links with
22
48 channels [27] and 72 channels (36 in each direction)
[28] have recently appeared on the market. Instead of
fiber ribbons, fiber imaging guides (FIGs) with thousands of pixels can be used [29]. In the system described
in [30], both a 14000-pixel FIG and a 3500-pixel FIG
were coupled to an 8 8 VCSEL (Vertical Cavity
Surface Emitting Laser) array in different set-ups. Over
short communication distances, the fiber ribbon can
even be replaced by some optical elements to obtain
free-space communication. A 256-channel bidirectional
intra-PCB (Printed Circuit Board) free-space interconnection system is described in [31].
3.2 Bit-Parallel WDM Links
A bit-parallel WDM link is another way to
synchronously transmit on several channels in parallel. In
this way we get bit-parallel byte-serial transmission [32].
However, compensation for bit skew caused by group
delay dispersion, where different wavelength channels
travel at different speeds in the fiber, may be needed in
these systems [33].
A dedicated wavelength in a bit-parallel WDM
link can be used for clock information. Wavelengths, or
fibers in a fiber-ribbon cable, can also be dedicated to
other purposes such as frame synchronization and flow
control. Significantly higher bandwidth distance products can be achieved when using bit-parallel WDM over
dispersion-shifted fiber instead of fiber-ribbons [34]. If,
however, there is only communication over shorter
distances, like a few meters, the bandwidth distance
product is not necessarily a limiting factor. Transmission
experiments with an array of eight pie-shaped VCSELs
arranged in a circular area with a diameter of 60 m, to
match the core of a multimode fiber, have been reported
[35]. Other work on the integration of components
for short distance WDM links has been reported,
e.g. a 4 2.5 Gbit/s transceiver with integrated splitter,
combiner, filters etc. [36].
4 Fiber-Optic Interconnection Networks
Fiber-optic network architectures, especially passive
ones, are discussed below. First, different basic passive
fiber-optic network architectures are described. WDM
star, WDM ring, and AWG networks are then respectively discussed. Fiber-ribbon ring networks are presented last.
4.1 Passive Fiber-Optic Networks
In an all-optical network, the data stream remains
in the optical form all the way from the transmitter to
the receiver. Three basic architectures for all-optical
multi-access networks are the ring, the bus and the
star (see Figure 5). These will be discussed below.
Most work on passive optical networks has focused on
LANs or similar structures but they can also be used as
Optical Networks Magazine July/August 2003
5138454
6/23/03
10:44 AM
Page 23
Figure 5: Passive optical network architectures: (a) ring, (b) (dual)
bus, and (c) star.
All of the three basic network architectures offer
different advantages. The ring has the least amount of
fibers, a bus network’s medium access protocol can utilize
the linear ordering of the nodes, and the star has the best
power budget in practice. Star networks are the most
popular, however, judging from the number of published
reports.
While passive optical networks that offer only one
shared channel are not promising alternatives, they form
the basis for more powerful networks using WDM.
These networks can be promising solutions as switch
fabrics in some kinds of data and telecommunication
equipment.
4.2 WDM star
substitutes for switch fabrics in data and telecommunication equipment. It should be noted that if one of
these networks is used in an OXC or the like as a switch
fabric, the signal is normally converted to electrical and
back to optical form at the entrance and the exit of the
switch fabric.
An all-optical multiple-access ring network differs
from a traditional ring network in the sense that all other
nodes can be reached in a single hop without any intermediate optoelectronic conversion. Only a fraction of
the optical power contained in the bypassing fiber is
tapped to the receiver, which gives all nodes the
opportunity to read the message. In other words, multicast or broadcast is possible. Outgoing messages are
inserted into the ring. A node removes its own messages
from the ring after one round.
In an optical bus, the light travels only in one direction, making two buses necessary (upper and lower),
one for each direction (higher or lower node index of
destination nodes). This kind of bus architecture
is called dual bus. The disadvantage of the dual bus is
that two transceivers are needed in each node. This
is avoided in the folded bus, where the two buses are
connected with a wrap-around connection at one end of
the buses (see Figure 6). In the folded bus, transmitters
and receivers are connected to the upper and lower bus,
respectively.
In a star network, the incoming light waves from all
nodes are combined and uniformly distributed back to
the nodes. In other words, the optical power contained in
the middle of the star is equally divided between all nodes.
Figure 6: Folded fiber-optic bus.
Optical Networks Magazine July/August 2003
By using WDM, multiple wavelength channels can
carry data simultaneously in the network. Figure 7 shows
an example of a WDM star network configuration. Each
node transmits on a wavelength unique to the node, while
the receiver can listen to an arbitrary wavelength [37].
One can say that this kind of network architecture implements a distributed crossbar. The flexibility is hence high,
and multicast and single-destination traffic can co-exist.
WDM star networks have been proposed especially for
internal use in packet switches [38-41].
Tunable components with tuning latencies in the
order of a nanosecond have been reported, but they often
have a limited tuning range [42]. At the expense of longer
tuning latencies, however, components with a broader
tuning range can be used [43]. Such components can be
used to achieve a cheaper network in systems where much
of the communication patterns remain constant for a
longer period.
Complete removal of the ability to tune in a WDM
star network gives a multi-hop network [44]. Each node
in a multi-hop network transmits and receives on one or
a few dedicated wavelengths. If a node does not have the
capability of sending on one of the receiver wavelengths
of the destination node, the traffic must pass one or
several intermediate nodes. The wavelengths can, for
example, be chosen to get a perfect-shuffle network [45].
A network in which several topologies, like a ring and
Figure 7: WDM star network.
23
5138454
6/23/03
10:44 AM
Page 24
Figure 8: WDM star multi-hop network.
Figure 10: AWG network.
a stand-alone network component [48] and as a component in routers [4].
4.5 Fiber-ribbon pipeline ring network
Figure 9: Multi-hop topology.
a hypercube, are embedded can also be chosen. An example of a multi-hop network is shown in Figure 8. This
configuration of wavelength assignments corresponds to
the topology shown in Figure 9.
Bit-parallel transfer can be utilized when fiber-ribbon
links are used to connect the nodes in a point-to-point
linked ring network. In such a network, one of the fibers in
each ribbon is dedicated to carrying the clock signal thus
making clock-recovery circuits unnecessary in the receivers.
Other fibers can be utilized for frame synchronization and
other control purposes. Figure 11 shows how a ring network is used as a switch fabric.
As seen in Figure 12, aggregated throughputs higher
than 1 can be obtained in ring networks with support for
spatial bandwidth reuse, sometimes called pipeline rings.
This feature can be effectively used when most of the
communication is to the nearest downstream neighbor.
Two fiber-ribbon pipeline ring networks have recently been
reported [49]. The first has support for circuit switching
4.3 WDM ring
A WDM ring network utilizes ADMs (Add Drop
Multiplexers) in all nodes to insert to, listen to and
remove wavelength channels from the ring. In the
WDMA ring network described in [46], each node is
assigned a node-unique wavelength on which to transmit. The other nodes can then tune in an arbitrary
channel on which to listen. This configuration is logically the same as that of the WDM star network with
fixed transmitters and tunable receivers. The distributed
crossbar again gives good performance for general
communication patterns.
Figure 11: Ring network as switch-fabric.
4.4 AWG networks
AWG (Arrayed Waveguide Grating) networks are
related to WDM star networks, but here the passive
optical star is exchanged with an AWG [47]. The AWG
routes wavelengths such that spatial reuse of wavelengths
is possible, as seen in Figure 10. Only N wavelengths (A,
B, C, and D,) are needed to get a fully connected N N network with N 2 optical channels (A1, A2, . . .
D4). The AWGs have been proposed for use both as
24
Figure 12: Example of spatial bandwidth reuse. Node M sends to
Node 1 at the same time that Node 1 sends to Node 2 and Node 2
sends a multicast packet to Nodes 3, 4, and 5.
Optical Networks Magazine July/August 2003
5138454
6/23/03
10:44 AM
Page 25
Figure 13: Control channel based network built up with fiberribbon point-to-point links.
on 8 1 fibers (data and clock) and packet switching on
an additional fiber. The second network is more flexible
and has support for packet switching on 8 1 fibers and
uses a tenth fiber for control packets (see Figure 13). The
control packets carry MAC information for the collisionless MAC protocol with support for slot reserving. Slot
reserving can be used to get RTVCs (Real-Time Virtual
Channels) for which guaranteed bandwidth and a worstcase latency are specified (compare with circuit switching).
The fiber-ribbon ring network can offer rather high
throughputs due to the aggregated bandwidth of a fiberribbon cable, especially when there is a great deal of nearest
downstream neighbor communication. However, the
spatial reuse will probably be limited if used in data
or telecommunication equipment owing to more evenly
distributed destinations as compared to the parallel computing systems for which the network was first proposed.
Another fiber-ribbon ring network is the PONI
network (formerly USC POLO), which is proposed for
use in clusters of workstations and similar systems
[50,51]. Integrated circuits have been developed for the
network, and tests have been done [52,53].
changed into an even more compact variant of the fiberribbon ring network. In addition, there is a clock that distributes clock signals to the four nodes via equal length
fibers to keep the clock signals in phase. In other words,
a fiber-optic clock distribution network and a data network are integrated into one system. If one foil is placed
on each PCB in a rack, they can be passively connected to
each other via fiber-ribbon cables or a backplane foil.
Using polymer waveguides instead of fibers offers advantages such as the possibility of integrating splitters and
combiners into the foil and the potential for more cost
effective mass production [54].
Another way is to follow the proposed use of an array
of passive optical stars to connect processor boards in a
multiprocessor system via fiber-ribbon links, for which
experiments with 6 700 Mbit/s fiber-ribbon links were
done (see Figure 15) [57]. Of course, the processors
boards can be exchanged with transceiver cards and/or
switch cards. As indicated above, such a configuration can
be integrated by the use of polymer waveguides. The
power budget can, however, be a limiting factor to the
number of nodes and/or the distance. Advantages are
simple hardware owing to bit-parallel transmission, like
other fiber-ribbon solutions, and the broadcast nature, but
the star array can become a bottleneck as in all bus-like
systems. In a similar system, the star array is exchanged
with a chip with optoelectronics that has one incoming
ribbon from each node and one output ribbon [58]. The
output ribbon is coupled to an array of 1 N couplers so
5 Integrated Optical Interconnection
Systems
This section describes three types of more or
less purely integrated optical interconnection systems.
This type of interconnection system can be used to interconnect electronic switch chips or I/O interfaces. Integrated fiber and waveguide solutions, planar free space
optics, and free space optical backplanes are discussed.
Figure 14: A foil of fibers connects four nodes and distributes
clock signals to them.
5.1 Integrated fiber and waveguide
solutions
Fibers or other kinds of waveguides (hereafter commonly denoted as channels) can be integrated to form a
more or less compact system of channels. Fibers can be
laminated to form a foil of channels for use as intra-PCB
or back-plane interconnection systems [54-56]. Fiberribbon connectors are applied to fiber end-points of the
foil. An example is shown in Figure 14, where four nodes
are connected in a ring topology. The medium is simply
Optical Networks Magazine July/August 2003
Figure 15: An array of passive optical stars connects a number of
nodes via fiber-ribbon cables.
25
5138454
6/23/03
10:44 AM
Page 26
that each node has a ribbon connected to its receiver. The
chip couples the incoming traffic together in a way that
simulates a bus. At contention, the chip can temporarily
store packets.
Other similar systems include the integration of
fibers into a PCB for the purpose of clock distribution [59]. Distribution to up to 128 nodes was demonstrated. The fibers are laminated on one side of the
PCB, while integrated circuits are placed on the reverse
side. The end section of each fiber is bent 90 degrees to
lead the light through a so called via hole to the reverse
side of the PCB.
5.2 Planar free space optics
Placing electronic chips, including optoelectronic devices, and optical elements on a substrate where light beams
can travel gives a planar free space system (Figure 16)
[60-61]. Electronic chips are placed in a 2-dimensional
plane, while light beams travel in a 3-dimensional space. In
this way, optical systems can be integrated monolithically,
which gives compact, stable and potentially inexpensive systems [61]. A flexible network can be obtained by using spatial light modulators to dynamically direct the optical beams.
Using only fixed interconnection patterns and electronic
switching can give shorter switch times, however. A planar
free space optical crossbar switch has been reported [62].
5.3 Free space optical backplanes
Several different optical backplanes have been proposed, three of which are discussed below. As shown in Figure 17a, using planar free space optics is one means of
transporting optical signals between PCBs. Holographic
gratings can be used to insert/extract the optical signals
to/from the waveguide, which may be a glass substrate [63].
Several beams or bus lines can be used so each arrow in the
figure represents several parallel beams [64].
In the system shown in Figure 17b, 2-dimensional
arrays of optical beams (typically 10 000) link neighboring
PCBs together in a point-to-point fashion [65]. Smart pixel
arrays then act as intelligent routers that can, e.g., bypass
data or perform data extraction operations where some data
pass to the local PCB and some data are retransmitted to
the next PCB [66]. Each smart pixel array can typically
contain 1 000 smart pixels arranged in a 2-dimensional
array, where each pixel has a receiver, a transmitter, and a
simple processing unit. One way to configure the system is
Figure 16: Example of a planar free space system. The beam direction is steered by the optical element on its way between two chips.
26
Figure 17: Optical backplane configurations: (a) with planar free
space optics, (b) with smart pixel arrays, and (c) with a mirror.
to connect the smart pixel arrays in a ring, where the ring
can be reconfigured to embed other topologies [67].
The configuration shown in Figure 17c is similar to
the optical backplane based on planar free space interconnects, but the waveguide is replaced here by a mirror [68].
An optical beam leaving a transmitter is simply bounced
once on the mirror before it arrives at the receiver.
A regeneration of the optical signal might be needed on
the way from the source to the final destination.
Of the three types of optical backplanes discussed,
the one with smart pixel arrays seems to be the most
powerful. On the other hand, a simple passive optical
backplane may have other advantages. Other optical
backplanes proposed include a bus where optical signals
can pass through transparent photo detectors or be
modulated by spatial light modulators [69].
6 Optical and Optoelectronic
Switch-Fabrics
This section introduces optical and optoelectronic
switch-fabrics. The combination of optical interconnections and electronic crossbars is treated first and a discussion is then presented of WDM/SDM switches.
6.1 Optical interconnections and electronic
crossbars
The switch itself can be modified to increase performance or packing density. A single-chip switch core
where fiber-ribbons are coupled directly to optoelectronic devices on the chip is possible [70]. Attaching 32
incoming and 32 outgoing fiber-ribbons with 800
Mbit/s per fiber translates to an aggregated bandwidth of
204 Gbit/s through the switch when there are eight data
fibers per link.
Optical Networks Magazine July/August 2003
5138454
6/23/03
10:44 AM
Page 27
A 16 16 crossbar switch chip with integrated optoelectronic I/O was implemented for switching packets
transferred using bit-parallel WDM [71]. Each node has
two single-mode fibers coupled to the switch, one for input
and one for output. Another switch system supporting an
arbitrary topology uses 8 8 crossbars, where a 12-channel
fiber-ribbon link is connected to each port [72]. The aggregated throughput of a single switch is 64 Gbit/s. Multistage
networks with free-space interconnections between electronic crossbars have also been reported [10,73,74]. Chips
or multi-chip-modules with surface-attached optical I/O
are stacked to obtain a 3-dimensional multistage system.
6.2 WDM/SDM switches
The architecture with optical interconnections and
electronic crossbars is flexible and powerful. In addition to
simply in the I/O interface or as optical interconnections
between electronic crossbars, optics and optoelectronics
can also be used internally in a switch fabric, however.
A broad spectrum of solutions has been proposed, and
some examples are given below.
SDM (Space Division Multiplexing) switches [13,7577] and WDM switches (with, e.g. wavelength converters
and wavelength selective components) [78–79] can be
used both as stand-alone switches and as building
components in larger switch fabrics [80]. As an example, a
Banyan multistage network built of 2 2 LiNbO3 switch
elements has been described [81-82], while a crossbar
equivalent system is described in [83]. Another multistage
network uses both WDM and SDM switches but in
different stages [84]. A multistage network can also be
implemented using chips with processors placed on a
2-dimensional plane [85]. The processors then communicate with each other by a mirror that bounces the beam
back to the plane but to another processor. Switching is
made on the chips while each pass between two switch
stages corresponds to a bounce on the mirror.
A multistage switch incorporating both electrical
and optical switching, but in different stages, has also
been reported [86]. Some work has focused on the communication between stages. An example is perfect shuffle
with lenses and prisms [14]. Switch times for SDM
switches in the order of 1 ns have been reported [77],
while some SDM switches have switch times in the order
of 1 ms [87]. A switch can be placed on a dedicated board
in a cabinet and be connected to processor boards or line
cards via fibers or an optical backplane [88].
A system that implements a distributed crossbar, or a
fully connected system, connecting N nodes with only passive optics between the transmitters and receivers has been
demonstrated [30]. All optical channels turned on from a
transmitter’s 2-dimensional 2N 2N VCSEL array
are inserted into a fiber image guide. The fiber image guides
from all transmitters end at a central free space system with
lenses. The lenses are arranged such that the light from each
Optical Networks Magazine July/August 2003
Figure 18: 2-dimensional optical MEMS switch.
VCSEL pixel in a VCSEL array is focused on a single spot,
together with the corresponding pixels in all other arrays.
This gives N spots at which each is focused into a single
fiber leading to a receiver. Hence, selecting a pixel in a
VCSEL array to be turned on corresponds to addressing a
destination node. More solutions with free-space connectivity, the fastest with nanoseconds switching, are found in [9].
Wavelength converters are important components in
many WDM switches. A way of building fast wavelength
converters is to have conversion first to the electrical
domain and then back to the optical domain but on
another wavelength. However, this is not a valid solution
if an all-optical switch is desired.
A great deal of attention has recently been paid to
using MEMS (microelectromechanical systems) technology
to build all-optical SDM switches [89]. As reported in
[90-91], an array of electrically controlled mirrors can be
used to build an 8 8 non-blocking OXC (see Figure 18).
The optical beams move in 2-dimensional space where each
mirror is controlled so that it is either in the down position
(no beam bounce) or the up position (beam bounce). The
disadvantage of the 2-dimensional MEMS switch is that N 2
mirrors are needed in an N N OXC. A more scalable
solution is to let the beams travel in 3-dimensional space
and to use mirrors for which the angle can be controlled
in two axis. In an N N OXC, two arrays of size
2N 2N are used. Each beam will first bounce on
an input-specific mirror in the first array and then on an
output-specific mirror in the second array. Lucent Technologies has already announced 256 256 OXCs based on
the 3-dimensional MEMS technology for release on the
market [92]. Due to the relatively low loss that can be
achieved in MEMS switches, multistage MEMS switches
are also possible [93]. In this way, rather large optically
transparent switches can be built. In addition to pure SDM
switches, the MEMS technology can be used in equipment
for wavelength routing networks, consisting of wavelength
splitters, a MEMS SDM switch and wavelength combiners.
More information on all-optical switching is found in [94].
7 Discussion
When designing interconnection fabrics for use in
communication equipment one must consider the characteristics of the specific equipment. For example, the
27
5138454
6/23/03
10:44 AM
Page 28
interconnection network must have satisfactory blocking
property. A blocking network for use in a true or burstswitching packet-switch might be good enough if its
capacity well exceeds what is required on average. Splitting packets into fixed-sized cells might also ease the use
of a blocking network if the network configuration and
transfer of cells are synchronized to occur at regular intervals. If the interruption of a long-duration bit-stream is
prohibited, however, even a rearrangeable network might
be unacceptable.
Another design issue with impact on the latency and
the bandwidth is whether or to what extent an interconnection network includes delaying elements and elements with
bandwidth bottlenecks. Examples of elements delaying the
data are components that convert the signal between an
electrical and optical form, long paths with propagation
delays, processing delays and queuing delays. Examples of
possible bandwidth bottlenecks are low-quality optoelectronic components, electronic paths like in optoelectronic
wavelength converters, protocol processing, queue bandwidth and optical signal degradation. It should, however, be
noted that the bandwidth or throughput is not directly
Switch Time
fast
Optical
Transparency
internally
WDM ring
fast
Fiber-ribbon pipeline ring
limited by delays as experienced in a computer, where a
memory read delay can place the computer in an idle state.
Instead, the low delay through optoelectronic components
can often be neglected when studying the total delay of
crossing cities or countries. The delay through network
equipment typically becomes considerable first when
accounting queuing delays at network congestion.
The interconnection architectures reviewed are summarized in Table 1 with remarks on their suitability in different aspects in data and telecommunication equipment.
Switch time is marked as slow (ms), medium (submicrosecond) or fast (ns) in the table depending on the
suitability of adoption in OXCs, packet switches with
burst switching or true packet switches, respectively.
Optical transparency is valuable for obtaining protocol-independent OXCs. The MEMS technology is promising for such systems, at least as long as the requirements
on switch times are moderate. Multistage networks using
MEMS technology can be especially good alternatives
because of their scalability. All-optical packet switches,
however, will probably not become a mature technology
in the near future. One must also think of the flexibility
Scalability
poor to
medium
Blocking
non-blocking
Notes
At slow switching, the
component’s tuning time
requirements are relaxed.
internally
poor
non-blocking
At slow switching, the
component’s tuning time
requirements are relaxed.
fast
no
poor
blocking
Can be compared with a
high performance bus but
with spatial bandwidth reuse.
Free space optical backplanes
varies a lot
internally
for some
systems
varies a lot
varies a lot
Scalable if there are many
I/O channels on one card.
Switching and line cards
can be mixed.
Nonblocking MEMS system
slow
yes
medium
(or better)
non
Multistage MEMS system
slow
yes
good
topology
dependent
Scalable and optically
transparent for systems
with relaxed switching time
requirements.
WDM/SDM switches
slow
(or better)
yes (for some
systems)
varies a lot
topology
dependent
Optical interconnections
and electronic crossbar
fast
no
poor to
medium
non-blocking
Planar free space optics
fast if no
SLMs
no
good
topology
dependent
For optical transparency,
only all-optical wavelength
conversion is allowed.
Many optoelectronic I/O
channels can be integrated on
a switching chip/module.
Chips are placed only in two
dimensions. Promising in terms
of assembly.
WDM star distributed
crossbar
Table 1: Summarizing evaluation of reviewed interconnection architectures.
28
Optical Networks Magazine July/August 2003
5138454
6/23/03
10:44 AM
Page 29
and power of electronic switches and of electronic processors to control the switches.
Scalability is desirable to be able to build equipment
with many in/output ports. We state the scalability as
poor if only tens of ports are realistic, medium for
hundred to a few hundred ports and good for a thousand
ports or more. Different multistage networks and free
space optical backplanes with a high density of optical
channels seem to be good candidates from a scalability
point of view. When building smaller systems instead the
WDM star distributed crossbar with its passive optical
star can be a good and simple alternative.
If an interconnection network implements a true
crossbar (like the WDM star distributed crossbar) it is
non-blocking, while whether a network is blocking or
non-blocking is topology dependent for many of the
architectures reviewed. It should however be noted that
the fiber-ribbon pipeline ring network is blocking. On
the other hand, the increasingly good price/performance
ratio for fiber-ribbon links indicates a great success potential for interconnection systems using fiber-ribbon links.
It might be possible to build larger switch-fabrics
with high transmission capacities using optics inside a
switch. The suitability of the different free space systems
depends a great deal on the more detailed configurations
of the systems. For example, planar free space systems can
be arranged in arbitrary topologies.
Integrated fiber and waveguide solutions make
possible the building of compact systems, especially for
networks that use fiber-ribbons. The same reasoning about
compactness can be argued for free space systems. Optical
backplanes may earn their success from their similarities
with current rack-based systems, while future planar freespace systems might give the possibility to integrate optics
and electronics in a compact and easy to assemble way.
For a more application-oriented comparison, Table 2
summarizes about the suitability of the different interconnection architectures for OXCs, packet switches with
burst switching and true packet switches. The switch
times of the different interconnection networks have
the largest influence on this suitability. In addition to the
pure interconnection demands, one must consider how
OXCs
Not suitable when optical
transparency is needed
Packet switches
with burst switching
Good for moderate
number of ports
True packet-switches
Good for moderate number
of ports if tuning time is low
WDM ring
Not suitable when optical
transparency is needed
Good for small number
of ports
Good for small number of
ports if tuning time is low
Fiber-ribbon pipeline ring
Not suitable when optical
transparency is needed
Good for small systems if
capacity is enough to deal
with blocking
Good for small systems
if capacity is enough to deal
with blocking
Free space optical backplanes
A passive optical backplane
can connect other optical
components placed on
inserted boards
Passive backplanes can connect
boards while smart-pixel based
backplanes even can participate
in the routing decisions
Passive backplanes can connect
boards while smart-pixel based
backplanes even can participate
in the routing decisions
Nonblocking MEMS system
2D MEMS is good for small
systems, while 3D MEMS
can support rather
many ports
Not suitable because of
long switch times
Not suitable because of
long switch times
Multistage MEMS system
High number of ports is
possible, when power
budget is acceptable
Not suitable because of
long switch times
Not suitable because of long
switch times
WDM/SDM switches
A wide area of solutions
exists, where wavelength
routing functionality often
is integrated
Typically not targeted for
switching-times short enough
for burst switching
Typically not targeted for
switching-times short
enough for packet switching
Optical interconnections
and electronic crossbar
Not suitable when optical
transparency is needed
Flexible solution for both
small systems (single crossbar)
and large systems (multistage
or similar)
Flexible solution for both
small systems (single crossbar)
and large systems (multistage
or similar)
Planar free space optics
Integration of reconfigurable
all-optical devices must
be investigated
Can interconnect electronic
crossbars in desired pattern
Can interconnect electronic
crossbars in desired pattern
WDM star distributed
crossbar
Table 2: Application suitability remarks.
Optical Networks Magazine July/August 2003
29
5138454
6/23/03
10:44 AM
Page 30
buffering and routing intelligence are included in the communication equipment. Such components or functions
can both be integrated into the interconnection fabric and
at the entrances and exits of the fabric. A more detailed
study of this is out of the scope of this paper, however.
8 Summary
In this paper, we have surveyed a number of interconnection architectures and technologies for use in
interconnection fabrics in communication equipment.
Some of the technologies can be used in OXCs, burstswitching packet-switches and true packet-switches,
while a MEMS-based interconnection fabric is limited to
use in an OXC because of its long switch-time. On the
other hand, the MEMS technology offers a good choice
when requiring optical transparency. In packet switches
instead, including electronic crossbars or other electronics
can offer required intelligence and flexibility for protocol
processing.
[9]
[10]
[11]
[12]
9 References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
30
F. Callegati, A.C. Cankaya, Y. Xiong, and M. Vandenhoute, “Design issues of optical IP routers for Internet
backbone applications,” IEEE Communications
Magazine, vol. 37, no. 12, pp. 124-128, Dec. 1999.
J. S. Turner, “Terabit burst switching,” Journal of
High Speed Networks, vol. 8, no. 1, pp. 3-16, 1999.
D. J. Blumenthal, A. Carena, L. Rau, V. Curri, S.
Humphries, “WDM optical IP tag switching with
packet-rate wavelength conversion and subcarrier
multiplexed addressing,” Optical Fiber Communication Conference, OFC’99, pp. 162-164, 1999.
H. J. Chao and T.-S. Wang, “An optical interconnection network for terabit IP routers,” Journal of Lightwave Technology, vol. 18, no. 12, pp. 2095-2112,
Dec. 2000.
G. Hjálmtýsson, J. Yates, S. Chaudhuri, A. Greenberg, “Smart routers–simple optics: an architecture
for the optical internet,” Journal of Lightwave Technology, vol. 18, no. 12, pp. 1880-1891, Dec. 2000.
T.-S. Wang and S. Dixit, “A scalable and high
capacity all-optical packet switch: design, analysis,
and control,” Photonic Network Communications,
vol. 3, no. 1/2, pp. 101-110, 2001.
D. Chiaroni, B. Lavigne, L. Hamon, A. Jourdan,
F. Dorgeuille, C. Janz, E. Grard, M. Renaud, R.
Bauknecht, C. Graf, H.P Schneibel, and H. Melchior,
“Experimental validation of an all-optical network
based on 160 Gbit/s throughput packet switching
nodes,” Proc. 24th European Conference on Optical
Communication, Madrid, Spain, Sept. 20-24, 1998,
vol. 1, pp. 573-574.
M. Jonsson, “Fiber-optic interconnection networks
for signal processing applications,” 4th International Workshop on Embedded HPC Systems and
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
Applications (EHPC’99) held in conjunction with the
13th International Parallel Processing Symposium &
10th Symposium on Parallel and Distributed Processing (IPPS/SPDP ‘99), San Juan, Puerto Rico, Apr.
16, 1999. Published in Lecture Notes in Computer
Science. vol. 1586, Springer Verlag, pp. 1374-1385,
1999, ISBN 3-540-65831-9.
M. M. Eshaghian, “Parallel algorithms for image
processing on OMC,” IEEE Transactions on Computers, vol. 40, no. 7, pp. 827-833, July 1991.
F. E. Kiamilev, P. Marchand, A. V. Krishnamoorthy,
S. C. Esener, and S. H. Lee, “Performance comparison between optoelectronic and VLSI multistage
interconnection networks,” Journal of Lightwave Technology, vol. 9, no. 12, pp. 1674-1692, Dec. 1991.
T. M. Pinkston and C. Kuznia, “Smart-pixel-based
network interface chip,” Applied Optics, vol. 36,
no. 20, pp. 4871-4880, July 10, 1997.
J. W. Goodman, F. I. Leonberger, S.-Y. Kung, and
R. A. Athale, “Optical interconnections for VLSI
systems,” Proceedings of the IEEE, vol. 72, no. 7,
pp. 850-866, July 1984.
A. A. Sawchuk, B. K. Jenkins, C. S. Raghavendra,
and A. Varma, “Optical crossbar networks,” Computer, vol. 20, no. 6, pp. 50-60, June 1987.
A. W. Lohmann, W. Stork, and G. Stucke, “Optical
perfect shuffle,” Applied Optics, vol. 25, no. 10,
pp. 1530-1531, May 15, 1986.
J. W. Goodman, “Fan-in and fan-out with optical
interconnections,” Optica Acta, vol. 32, no. 12,
pp. 1489-1496, 1985.
H. S. Hinton, “Architectural considerations for photonic
switching networks,” IEEE Journal of Selected Areas in
Communications, vol. 6, no. 7, pp. 1209-1226, 1988.
http://www.velio.com/.
D. B. Schwartz, K. Y. Chun, N. Choi, D. Diaz,
S. Planer, G. Raskin, and S. G. Shook,
“OPTOBUS™ I: performance of a 4 Gb/s optical
interconnect,” Proc. Massively Parallel Processing
using Optical Interconnections (MPPOI’96), Maui,
HI, USA, Oct. 27-29, 1996, pp. 256-263.
C.-L. Jiang, D. J. Brown, E. H. Mueller, A. E. Plotts,
E. Cornejo, S. O’Neill, A. J. Heiney, and B. H. Reysen,
“LED based parallel optical links,” Proc. LEOS’95,
San Francisco, CA, USA, Oct. 30-Nov. 2, 1995,
vol. 1, pp. 224-225.
F. A. P. Tooley, “Optically interconnected electronics
– challanges and choices,” Proc. Massively Parallel
Processing using Optical Interconnections (MPPOI’96),
Maui, HI, USA, Oct. 27-29, 1996, pp. 138-145.
T. Kurokawa, S. Matso, T. Nakahara, K. Tateno,
Y. Ohiso, A. Wakatsuki, and H. Tsuda, “Design
approaches for VCSEL’s and VCSEL-based smart
pixels toward parallel optoelectronic processing systems,” Applied Optics, vol. 37, no. 2, pp. 194-204,
Jan. 10, 1998.
Optical Networks Magazine July/August 2003
5138454
6/23/03
10:44 AM
Page 31
[22] A. P. Kanjamala and A. F. J. Levi, “Sub-picosecond
skew in multimode fiber ribbon for synchronous
data transmission,” Electronics Letters, vol. 31,
pp. 1376-1377, 1995.
[23] Y.-M. Wong et al., “Technology development of a
high-density 32-channel 16-Gb/s optical data
link for optical interconnection applications for
the optoelectronic technology consortium
(OETC),” Journal of Lightwave Technology, vol.
13, no. 6, pp. 995-1016, June 1995.
[24] T. Yoshikawa, S. Araki, K. Miyoshi, Y. Suemura,
N. Henmi, T. Nagahori, H. Matsuoka, and T. Yokota,
“Skewless optical data-link subsystem for massively
parallel processors using 8 Gb/s 1.1 Gb/s MMF
array optical module,” IEEE Photonics Technology Letters,” vol. 9, no. 12, pp. 1625-1627, Dec. 1997.
[25] N. Fujimoto, A. Ishizuka, H. Rokugawa, and K. Mori,
“Skew-free parallel optical transmission systems,”
Journal of Lightwave Technology, vol. 16, no. 10,
pp. 1822-1831, Oct. 1998.
[26] K. Kasahara, “Optical interconnects speed up networks,” Photonics Spectra, vol. 32, no. 2, pp. 127-128,
1998.
[27] http://www.teraconnect.com/.
[28] http://www.xanoptix.com/.
[29] Y. Li, H. Kosaka, T. Wang, S. Kawai, and K. Kasahara, “Applications of fiber image guides to bit-parallel optical interconnections,” Optical Computing,
vol. 10, 1995 OSA Technical Digest Series, Salt Lake
City, Utah, Mar. 13–16, 1995, pp. 286-288.
[30] Y. Li, T. Wang, and S. Kawai, “Distributed crossbar
interconnects with vertical-cavity surface-emitting
laser-angle multiplexing and fiber image guides,”
Applied Optics, vol. 37, no. 2, pp. 254-263, Jan. 10,
1998.
[31] D. V. Plant, M. B. Venditti, E. Laprise, J. Faucher,
K. Razavi, M. Châteauneuf, A. G. Kirk, and
J. S. Ahearn, “256-channel bidirectional optical
interconnect using VCSELs and photodiodes on
CMOS,” Journal of Lightwave Technology, vol. 19,
no. 8, pp. 1093-1103, Aug. 2001.
[32] M. L. Loeb and G. R. Stilwell, “High-speed data
transmission on an optical fiber using a byte-wide
WDM system,” Journal of Lightwave Technology,
vol. 6, no. 8, pp. 1306-1311, Aug. 1988.
[33] G. Jeong and J. W. Goodman, “Long-distance
parallel data link using WDM transmission with
bit-skew compensation,” Journal of Lightwave Technology, vol. 14, no. 5, pp. 655-660, May 1996.
[34] L. Bergman, J. Morookian, and C. Yeh, “An all-optical
long-distance multi-Gbytes/s bit-parallel WDM
single-fiber link,” Journal of Lightwave Technology, vol.
16, no. 9, pp. 1577-1582, Sept. 1998.
[35] L. A. Coldren, E. R. Hegblom, Y. A. Akulova, J. Ko,
E. M. Strzelecka, and S. Y. Hu, “Vertical-cavity
lasers for parallel optical interconnects,” Proc. 5th
Optical Networks Magazine July/August 2003
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
International Conference on Massively Parallel Processing using Optical Interconnections (MPPOI’98), Las
Vegas, NV, USA, June 15-17, 1998, pp. 2-10.
L. B. Aronson, B. E. Lemoff, L. A. Buckman, and
D. W. Dolfi, “Low-cost multimode WDM for local
area networks up to 10 Gb/s,” IEEE Photonics
Technology Letters,” vol. 10, no. 10, pp. 1489-1491,
Oct. 1998.
M. Jonsson, A. Åhlander, M. Taveniku, and
B. Svensson, “Time-deterministic WDM star
network for massively parallel computing in radar
systems,” Proc. Massively Parallel Processing using
Optical Interconnections (MPPOI’96), Maui, HI,
USA, Oct. 27-29, 1996, pp. 85-93.
K. Y. Eng, “A photonic knockout switch for highspeed packet networks,” IEEE Journal on Selected Areas
in Communications, vol. 6, no. 7, pp. 1107-1116,
Aug. 1988.
C. A. Brackett, “On the capacity of multiwavelength optical-star packet switches,” IEEE LTS,
pp. 33-37, May 1991.
B. Li, Y. Qin, X.-R. Cao, and K. M. Sivalingam,
“Photonic packet switching: architectures and
performance,” Optical Networks Magazine, vol. 2,
no. 1, pp. 27-39, Jan./Feb. 2001.
D. Sadot and I. Elhanany, “Optical switching
speed requirements for Terabit/second packet over
WDM networks,” IEEE Photonics Technology Letters,” vol. 12, no. 4, pp. 440-442, Apr. 2000.
H. Kobrinski, M. P. Vecchi, E. L. Goldstein, and
R. M. Bulley, “Wavelength selection with nanosecond switching times using distributed-feedback
laser amplifiers,” Electronics Letters, vol. 24, no. 15,
pp. 969-971, July 21, 1988.
K.-W. Cheung, “Acoustooptic tunable filters in
narrowband WDM networks: system issues and
network applications,” IEEE Journal on Selected
Areas in Communications, vol. 8, no. 6, pp. 10151025, Aug. 1990.
B. Mukherjee, “WDM-based local lightwave networks part II: multihop systems,” IEEE Network,
pp. 20-32, July 1992.
A. S. Acampora and M. J. Karol, “An overview of
lightwave packet networks,” IEEE Network,
pp. 29-41, Jan. 1989.
M. I. Irshid and M. Kavehrad, “A fully transparent
fiber-optic ring architecture for WDM networks,”
Journal of Lightwave Technology, vol. 10, no. 1,
pp. 101-108, Jan. 1992.
M. Maier and A. Wolisz, “Demonstrating the
potential of arrayed-waveguide grating based singlehop WDM networks,” Optical Networks Magazine,
vol. 2, no. 5, pp. 75-85, Sept./Oct. 2001.
Y. Sakai, “Full-mesh wavelength-routing WDM
network based on arrayed-waveguide grating,” Proc.
LEOS’95, vol. 2, pp. 832-833, 2000.
31
5138454
6/23/03
10:44 AM
Page 32
[49] M. Jonsson, “Two fiber-ribbon ring networks for
parallel and distributed computing systems,” Optical
Engineering, vol. 37, no. 12, pp. 3196-3204, Dec. 1998.
[50] B. Raghavan, Y.-G. Kim, T.-Y. Chuang, B. Madhavan,
and A. F. J. Levi, “A Gbyte/s parallel fiber-optic
network interface for multimedia applications,”
IEEE Network, vol. 13, no. 1, pp. 20-28, Jan./Feb.
1999.
[51] B. J. Sano and A. F. J. Levi, “Networks for the
professional campus environment,” in Multimedia
Technology for Applications. B. Sheu and M. Ismail,
Eds., McGraw-Hill, Inc., pp. 413-427, 1998, ISBN
0-7803-1174-4.
[52] B. Sano, B. Madhavan, and A. F. J. Levi, “8 Gbps
CMOS interface for parallel fiber-optic interconnects,” Electronics Letters, vol. 32, pp. 2262-2263,
1996.
[53] POLO Technical Summary. University of Southern
California, Oct. 1997.
[54] P. Eriksen, K. Gustafsson, M. Niburg, G. Palmskog,
M. Robertsson, and K. Åkermark, “The Apollo
demonstrator – new low-cost technologies for optical
interconnects,” Ericsson Review, vol. 72, no. 2, 1995.
[55] M. Robertsson, K. Engberg, P. Eriksen, H.
Hesselbom, M. Niburg, and G. Palmskog, “Optical
interconnects in packaging for telecom applications,” Proc. of the 10th European Microelectronics
Conference, pp. 580-591, 1995.
[56] M. A. Shahid and W. R. Holland, “Flexible optical
backplane interconnections,” Proc. 3rd International Conference on Massively Parallel Processing
using Optical Interconnections (MPPOI’96), Maui,
HI, USA, Oct. 27-29, 1996, pp. 178-185.
[57] J. W. Parker, P. J. Ayliffe, T. V. Clapp, M. C. Geear,
P. M. Harrison, and R. G. Peall, “Multifibre bus for
rack-to-rack interconnects based on opto-hybrid
transmitter/receiver array pair,” Electronics Letters,
vol. 28, no. 8, pp. 801-803, April 9, 1992.
[58] P. Lukowicz, S. Sinzinger, K. Dunkel, and
H. D. Bauer, “Design of an opto-electronic VLSI/
parallel fiber bus,” Proc. Optics in Computing (OC’98),
Brugge, Belgium, June 17-20, 1998, pp. 289-292.
[59] Y. Li, J. Popelek, J.-K. Rhee, L. J. Wang, T. Wang,
and K. Shum, “Demonstration of fiber-based boardlevel optical clock distributions,” Proc. 5th International Conference on Massively Parallel Processing
using Optical Interconnections (MPPOI’98), Las
Vegas, NV, USA, June 15-17, 1998, pp. 224-228.
[60] J. Jahns, “Planar packaging of free-space optical
interconnects,” Proceedings of the IEEE, vol. 82, no.
11, pp. 1623-1631, Nov. 1994.
[61] J. Jahns, “Integrated free-space optical interconnects for chip-to-chip communications,” Proc. 5th
International Conference on Massively Parallel Processing using Optical Interconnections (MPPOI’98),
Las Vegas, NV, USA, June 15–17, 1998, pp. 20-23.
32
[62] S. Reinhorn, R. Oron, Y. Amitai, A. A. Friesem,
K. Vinokur, and N. Pilossof, “Planar optical
dynamic crossbar switch,” Optical Engineering, vol.
38, no. 8, pp. 1396-1401, Aug. 1999.
[63] C. Zhao, T.-H. Oh, and R. T. Chen, “General purpose bidirectional optical backplane: high-performance bus for multiprocessor systems,” Proc. 2nd
International Conference on Massively Parallel Processing using Optical Interconnections (MPPOI’95), San
Antonio, TX, USA, Oct 23-24, 1995, pp. 188-195.
[64] C. Zhao, J. Liu, and R. T. Chen, “Hybrid optoelectronic backplane bus for multiprocessor-based computing systems,” Proc. 3rd International Conference
on Massively Parallel Processing using Optical Interconnections (MPPOI’96), Maui, HI, USA, Oct.
27-29, 1996, pp. 313-320.
[65] T. H. Szymanski, “Intelligent optical backplanes,”
Optical Computing, vol. 10, 1995 OSA Technical
Digest Series, Salt Lake City, Utah, Mar. 13-16,
1995, pp. 11-13.
[66] B. Supmonchai and T. Szymanski, “High speed
VLSI concentrators for terabit intelligent optical
backplanes,” Proc. Optics in Computing (OC’98),
Brugge, Belgium, June 17-20, 1998, pp. 306-310.
[67] T. Szymanski and H. S. Hinton, “Design of a
terabit free-space photonic backplane for parallel
computing,” Proc. 2nd International Conference on
Massively Parallel Processing using Optical Interconnections (MPPOI’95), San Antonio, TX, USA, Oct
23-24, 1995, pp. 16-27.
[68] K. Hirabayashi, T. Yamamoto, and S. Hino,
“Optical backplane with free-space optical interconnections using tunable beam deflectors and a mirror
for bookshelf-assembled terabit per second class
asynchronous transfer mode switch,” Optical Engineering, vol. 37, no. 4, pp. 1332–1342, Apr. 1998.
[69] K. Hamanaka, “Otical bus interconnection system
using selfloc lenses,” Optics Letters, vol. 16, no. 6,
pp. 1222-1224, Aug. 15, 1991.
[70] T. H. Szymanski, A. Au, M. Lafrenière-Roula,
V. Tyan, B. Supmonchai, J. Wong, B. Zerrouk, and
S. T. Obenaus, “Terabit optical local area networks
for multiprocessing systems,” Applied Optics, vol.
37, no. 2, pp. 264-275, Jan. 10, 1998.
[71] A. V. Krisnamoorthy et al., “The AMOEBA chip:
an optoelectronic switch for multiprocessor networking using dense-WDM,” Proc. 3rd International Conference on Massively Parallel Processing
using Optical Interconnections (MPPOI’96), Maui,
HI, USA, Oct. 27-29, 1996, pp. 94-100.
[72] S. Nishimura, T. Kudoh, H. Nishi, J. Yamamoto,
K. Harasawa, N. Matsudaira, S. Akutsu, and
H. Amano, “64-Gb/s highly reliable network
switch (RHiNET-2/SW) using parallel optical
interconnection,” Journal of Lightwave Technology,
vol. 18, no. 12, pp. 1620-1627, Dec. 2000.
Optical Networks Magazine July/August 2003
5138454
6/23/03
10:44 AM
Page 33
[73] F. B. McGormick et al., “Five-stage free-space optical
switching network with field-effect transistor selfelectro-optic-effect-device smart-pixel arrays,” Applied
Optics, vol. 33, no. 8, pp. 1601-1618, Mar. 1994.
[74] T. J. Cloonan, “Comparative study of optical
interconnection technologies for large asynchronous
transfer mode packet switching applications,” Optical Engineering, vol. 33, no. 5, pp. 1512–1523,
May 1994.
[75] T. Goh, M. Yasu, K. Hattori, A. Himeno,
M. Okuno, and Y. Ohmori, “Low-loss and highextinction-ratio silica-based strictly nonblocking
16 16 thermooptic matrix switch,” IEEE Photonics Technology Letters,” vol. 10, no. 6, pp. 810-812,
June 1998.
[76] D. A. B. Miller, D. S. Chemla, T. C. Damen,
A. C. Gossard, W. Wiegmann, T. H. Wood, and
C. A. Burrus, “Novel hybrid optically bistable
switch: the quantum well self-electro-optic effect
device,” Applied Physics Letters,” vol. 45, no. 1,
pp. 13-15, July 1, 1984.
[77] T. Kato, J. Sasaki, T. Shimoda, H. Hatakeyama,
T. Tamanuki, M. Yamaguchi, M. Kitamura, and
M. Itoh, “10 Gb/s photonic cell switching with
hybrid 4 4 optical matrix switch module on silica
based planar waveguide platform,” Optical Fiber
Communication Conference, OFC’98 Technical Digest, San Jose, CA, USA, Feb. 22–27, 1998, pp.
437–440.
[78] R. J. S. Pedersen, B. Mikkelsen, B. F. Jørgensen, M.
Nissov, K. E. Stubkjaer, K. Wünstel, K. Daub, E.
Lach, G. Laube, W. Idler, M. Schilling, P. Doussiere,
and F. Pommerau, “WDM cross-connect cascade
based on all-optical wavelength converters for routing and wavelength slot interchanging using a
reduced number of internal wavelengths,” Optical
Fiber Communication Conference, OFC’98 Technical
Digest, San Jose, CA, USA, Feb. 22-27, 1998,
pp. 58-59.
[79] R. Flipse, “Optical switches ease bandwidth crunch,”
EuroPhotonics, vol. 3, no. 5, pp. 44-45, Aug./Sept.
1998.
[80] J. H. Reif and A. Yoshida, “Free space optical
message routing for high performance parallel
computers,” Proc. Massively Parallel Processing using
Optical Interconnections (MPPOI’94), Cancun,
Mexico, Apr. 26-27, 1994, pp. 37-44.
[81] R. D. Chamberlain, M. A. Franklin, R. B. Krchnavek, and B. H. Baysal, “Design of an opticallyinterconnected multiprocessor,” Proc. 5th International Conference on Massively Parallel Processing using
Optical Interconnections (MPPOI’98), Las Vegas, NV,
USA, June 15-17, 1998, pp. 114-122.
[82] C. S. Baw, R. D. Chamberlain, M. A. Franklin, and
M. G. Wrighton, “The Gemini interconnect: data
path measurements and performance analysis,” Proc.
Optical Networks Magazine July/August 2003
[83]
[84]
[85]
[86]
[87]
[88]
[89]
[90]
[91]
[92]
[93]
[94]
6th International Conference on Parallel Interconnects
(PI’99), pp. 21-30, 1999.
P. Granestrand, B. Stoltz, L. Thylen, K. Bergvall,
W. Döldissen, H. Heinrich, and D. Hoffmann,
“Strictly nonblocking 8 x 8 integrated optical switch
matrix,” Electronics Letters, vol. 22, no. 15, pp. 816818, July 17, 1986.
S. Kawai, H. Kurita, and K. Kubota, “Design of
electro-photonic computer-networks with nonblocking and self-routing functions,” Optical Computing, vol. 10, 1995 OSA Technical Digest Series, Salt
Lake City, Utah, Mar. 13-16, 1995, pp. 263-265.
M. P. Christensen and M. W. Haney, “Two-bounce
free-space arbitrary interconnection architecture,”
Proc. Massively Parallel Processing using Optical
Interconnections (MPPOI’97), Montreal, Canada,
June 22-24, 1997, pp. 61-67.
C. Duan and C. W. Wilmsen, “Optoelectronic
ATM switch using VCSEL and smart detector
arrays,” Proc. Optics in Computing (OC’98), Brugge,
Belgium, June 17-20, 1998, pp. 103-106.
A. Tajima, N. Kitamura, S. Takahashi, S. Kitamura,
Y. Maeno, Y. Suemura, and N. Henmi, “10-Gb/s/
port gated divider passive combiner optical switch
with single-mode-to-multimode combiner,” IEEE
Photonics Technology Letters,” vol. 10, no. 1, pp. 162164, Jan. 1998.
Y. Maeno, A. Tajima, Y. Suemura, and N. Henmi,
“8.5 Gbit/s/port synchronous optical packet-switch,”
Proc. Massively Parallel Processing using Optical Interconnections (MPPOI’97), Montreal, Canada, June
22-24, 1997, pp. 114-119.
A. Neukermans and R. Ramaswami, “MEMS technology for optical networking applications,” IEEE
Communications Magazine, vol. 39, no. 1, pp. 62-69,
Jan. 2001.
L. Y. Lin, “Free-space micromachined opticalswitching technologies and architectures,” Optical
Fiber Communication Conference, OFC’99, vol. 2,
pp. 154-156, 1999.
C. Pu, L. Y. Lin, E. L. Goldstein, and R. W. Tkach,
“Client-configurable eight-channel optical add/drop
multiplexer using micromachining technology,”
IEEE Photonics Technology Letters,” vol. 12, no. 12,
pp. 1665, Dec. 2000.
M. Kenward, “Mirror magic ushers in the alloptical
network,” Fibre Systems, pp. 37-39, May 2000.
G. Shen, T. H. Cheng, S. K. Bose, C. Lu,
and T. Y. Chai, “Architectural design for multistage 2-D MEMS optical switches,” Journal of
Lightwave Technology, vol. 20, no. 2, pp. 178-187,
Feb. 2002.
A. Pattavina, M. Martinelli, G. Maier, and P. Boffi,
“Techniques and technologies towards all-optical
switching,” Optical Networks Magazine, vol. 1, no.
2, pp. 75-93, Apr. 2000.
33
5138454
6/23/03
10:44 AM
Page 34
Magnus Jonsson
[email protected]
Magnus Jonsson received his B.S. and M.S.
degrees in computer engineering from
Halmstad University, Sweden, in 1993
and 1994, respectively. He then obtained
the Licentiate of Technology and Ph.D.
degrees in computer engineering from
Chalmers University of Technology, Gothenburg, Sweden, in 1997
and 1999, respectively. From 1998 to March 2003, he was Associate
Professor of Data Communication at Halmstad University (acting
34
between 1998 and 2000). Since April 2003, he has been Professor of
Real-Time Computer Systems at Halmstad University. He has published about 35 scientific papers, most of them in the area of optical
communication and real-time communication. Most of his research is
targeted for embedded, industrial and parallel and distributed
computing and communication systems. Dr. Jonsson has served on the
program committees of International Workshop on Optical Networks,
IEEE International Workshop on Factory Communication Systems,
International Conference on Computer Science and Informatics, and
International Workshop on Embedded/Distributed HPC Systems
and Applications.
Optical Networks Magazine July/August 2003
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Related manuals

Download PDF

advertisement