Wavelength Swapping using Tunable Lasers for Fractional λ Switching

Abstract

Fractional lambda switching (Flambda;S) is a novel proposal for the management of all-optical networks with sub-wavelength provisioning capability. The unique characteristic of Flambda;S is the utilization of the UTC (coordinated universal time) for alignment and switching. Several central research issues are still open in Flambda;S and need to be formally defined and analyzed. Within the scope of this paper, we introduce three novel switch architecture designs that are based on the use of tunable lasers. As an important goal, we introduce the notion "scheduling feasibility" that measure the number of possible different scheduling between an input and output time-frames

Full text

Page 1

Abstract—Fractional Lambda Switching (Fλ λS) is a novel
proposal for the management of all-optical networks with sub-
wavelength provisioning capability. The unique characteristic of
Fλ λS is the utilization of the UTC (coordinated universal time) for
alignment and switching. Several central research issues are still
open in Fλ λS and need to be formally defined and analyzed.
Within the scope of this paper, we introduce three novel switch
architecture designs that are based on the use of tunable lasers.
As an important goal, we introduce the notion “scheduling
feasibility” that measure the number of possible different
scheduling between an input and output time-frames.
Index Terms—optical networks, sub-lambda switching, time-
driven switching, tunable laser, scheduling.
I. INTRODUCTION
ulti-wavelength optical networks [1] have been the
subject of research for many years. The bandwidth
granularity in optical channel routed networks is the
bandwidth of the whole optical channel or lambda (λ), i.e., it
is only possible to allocate the whole optical channel capacity
or nothing. Switching a whole optical channel is not
convenient, since each optical channel has a capacity ranging
from 2.5 Gb/s to 40 Gb/s and can accommodate a very large
number of conventional IP-traffic users. Thus, it is more
bandwidth efficient if an optical channel can be partitioned
into a number of sub-lambda or fractional lambda channels.
FλS capability is especially important in local and
metropolitan area networks (LAN/MAN), since the end-users’
traffic is very dynamic and requires only a fraction of the
optical channel. FλS solves the problem of sub-wavelength
traffic tricklets; however, to account for the highly dynamic
nature of Internet traffic, it has to be coupled with grooming
capabilities that dynamically multiplex IP level traffic into the
appropriate sub-wavelength
transmission. Grooming (multiplexing) the traffic from
multiple end-users is required in order to improve the
throughput and reduce the operation cost of optical networks.
The obvious solution is the implementation of asynchronous
IP-packet switching, but looking for all-optical networking,
asynchronous IP-packet switching is not suitable. Moreover,
future traffic will include a large portion of multimedia
distribution that is inherently synchronous. Consequently,
fractional lambda switching combined with grooming has the
potential to bring the benefits of optical networking in terms
of reliability and capacity to the end-users.
Recently, Optical Burst Switching (OBS) [2] was proposed
as a middle stage toward the realization of optical packet
“virtual container” for
switching. A burst accommodates a possibly large number of
packets. In OBS, control packets are forwarded in a control
channel to configure switching nodes before the arrival of
corresponding bursts, reducing the requirement of optical
buffers. Though OBS is interesting and some protocols were
defined for it [3][4], the behavior of burst switching as an
asynchronous switching system makes it hard to implement
and control the optical switching fabric even when the traffic
load is moderate. Besides, grooming traffic into bursts at
ingress nodes of OBS networks is another difficult issue. An
Optical Packet Switching (OPS) network [5] may be the
ultimate goal for all-optical networking. Unfortunately, the
technology to manufacture optical random access memory and
optical processing are not yet mature enough to realize OPS.
The contribution of this paper is the discussion of different
possible architectures for the realization of FλS nodes, their
complexity and their performance in terms of flexibility and
redundancy for scheduling and switching time-frames from an
input to an output port.
II. FλS - BASIC PRINCIPLES
A. FλS Timing Principle
Sub-lambda of fractional lambda switching was proposed as
an effort to realize highly scalable dynamic optical networking
[6][7][8], which requires minimum optical buffers. FλS has
the same general objectives as in OBS and OPS: obtaining
higher wavelength utilization, and realizing all-optical
networks. In FλS, a concept of common time reference using
UTC (Coordinated Universal Time) is introduced. A UTC
second is partitioned into a predefined number of time-frames
(see details in [6][7]). Time-frames can be viewed as virtual
containers for multiple IP packets that are switched, at every
FλS node, based on and coordinated by the UTC signal. As
shown in Fig.1, a group of k time-frames forms a time-cycle;
l contiguous time-cycles are grouped into a super cycle (in
Fig.1, k =1000, l=80).
Super-cycle
with 80k Time-frames
1 2
0
beginning
of a UTC second
1000
Time
Cycle0
f T
1 2
1000
Time
Cycle1
f T
1 2
1000
Time
Cycle 79
f T
CTR/UTC
1
beginning
of a UTC second
fT
f T

Fig. 1. Division of an UTC second in FλS [6].
Wavelength Swapping using Tunable Lasers
for Fractional λ Switching

Viet-Thang Nguyen*, M. Baldi**, R. Lo Cigno*, and Y. Ofek*, Senior Member, IEEE

*University of Trento, Dept. of Information and Telecommunications, Via Sommarive, 14 I-38050 Povo - Trento, ITALY
Tel. +39-04-6188.3954 – Email: {nguyen,locigno,ofek}@dit.unitn.it
** Politecnico di Torino - Dept. of Computer Engineering, Corso Duca degli Abruzzi, 24, 10129 Torino, ITALY
Email: mario.baldi@polito.it
M

Page 2

To enable FλS, time-frames are aligned at the inputs of
every FλS node before being switched. After alignment, the
delay between any pair of adjacent switch nodes is an integer
number of time-frames.
Another key element of FλS is the method of pipeline
forwarding. In FλS, a fractional lambda pipe (FλP) p is
defined as a predefined schedule for switching and forwarding
time-frames along a path of subsequent FλS-enabled nodes.
The FλP capacity is determined by the number time-frames
allocated in every time cycle (or super cycle) for the FλP p.
For example, for 10 Gb/s optical channel and k =1000, l=80 if
one time-frame is allocated in every time cycle or super cycle
the FλP capacity is 10 Mb/s or 125 kb/s, respectively.
B. FλS Forwarding Principle
FλS defines two possible types of forwarding, as shown in
Fig.2. The first is immediate forwarding (IF), upon the arrival
of each time-frame to an FλS node, the content of the time-
frame are scheduled to be “immediately” switched and
forwarded to the next node during the next time-frame. Hence,
the buffer that is required is bounded to one time-frame and
the end-to-end transmission delay is minimized.
The other type of packet forwarding is called non-
immediate forwarding (NIF). NIF requires buffers at FλS
nodes. Let us assume that, at each node, there is a buffer of b
time-frames at each input channel. The content of each time-
frame arriving to the FλS node can be buffered for
frames, 1b
≤≤
, before being forwarded to the next node.
NIF offers greater scheduling feasibility than IF. This
increasing flexibility that makes schedules feasible is one of
the main issues we discuss in this paper.
In FλS networks, a FλP is setup prior to the actual data
transmission. With an already-established FλP, the end-to-end
average delay is constant (i.e., accumulated buffer delay and
transmission delay) and no data is lost due to packet dropping.
C. Tunable Laser Principle – Wavelength Swapping
This work focuses on FλS with tunable lasers, since they
are available with high performances, e.g. a 16-channel 100-
GHz-spacing digitally tunable laser with 0.8 ns switching time
between channels has been experimented [9]. In general, the
way tunable lasers are used in this work is to change the
wavelength (color) of time-frames that contains IP packets
every FλS node. When wavelength converters will be
available they may replace the tunable lasers.
This operation can be viewed as wavelength swapping of
packets. Namely, packets are transmitted with λ1 over the first
optical link, then with λ2 over the second optical link and so
bk time-
bk

KK
K
K
kb-forwarding
Input
Output

Fig. 2. Illustration of IF and NIF in time domain.
on. The operation of swapping wavelength is equivalent to
label swapping. Obviously, as in label swapping, packets of
different connections (FλPs) should not have the same color
(label) when being transmitted over the same optical link and
having the same time index.
III. SCHEDULING FEASIBILITY OF FλS WITH TUNABLE LASER
We discuss and evaluate several optical switch
architectures, which combine FλS with tunable lasers. The
goal of each architecture is having the lowest possible
complexity and cost, while maintaining the performance in
terms of end-to-end blocking probability as low as possible.
The computation of the blocking probability is a hard task, but
we argue that it is a function of the scheduling feasibility as
defined below. Hence in this paper we use the later to quantify
the performance. Thus, the different switch architectures are
compared using: (1) the hardware complexity, (2) the
scheduling feasibility. The performance study of the blocking
probability and the relationship with the scheduling feasibility
is left for future study.
In order to give consistent and convenient descriptions of
the different switch architectures, the following notations are
used:
- C is the link capacity in terms of the number of wavelength
(colors) per optical link.
-
inout
NNN
==
is the number of input/output ports per
switch.
- r C N
=
is the connection ratio. For simplicity it is assumed
that r is integer.
-
( , )X i j denotes device j of type X (e.g., tunable laser TL
or star coupler SC ) of the in-port i .
-
- k denotes the size of time cycle in number of time-frames.
- h denotes the route length in number of hops.
Scheduling feasibility definition:* for a generic FλP the
scheduling feasibility is the number of distinct schedules that
are available using time and wavelength swapping. The
scheduling feasibility is function of: the forwarding method
(IF or NIF), k , h , C and
.
N
A schedule is a possible (not necessarily feasible)
assignment of resources (time-frames, optical channels…) to
build a FλP. A feasible schedule is not guaranteed to be
available at the time of FλS setup due to blocking (e.g.
switching fabric limitation, contention between multiple FλS
setups).
The switch architectures studied in this work have four
components:
1. WDM demultiplexers on the input side;
2. WDM multiplexers on the output side;
3. Tunable lasers that are connected to the WDM
demultiplexers;

* In this paper, we consider the schedule feasibility for scheduling 1 time-
frame FλPs. For the case of multiple time-frames scheduling, the IF scheme is
tractable using combinatorial mathematic. On the other hand, considering the
arbitrary NIF scheme, we have still not proven if it if it is tractable or not.
T R denotes the tuning range of a tunable laser.

Page 3

4. A connection network that connects the tunable lasers
with the WDM multiplexers at the outputs, which is in
essence what distinguish the various switch architectures
that are discussed in this paper.
In this work we study the following switch architectures:
- Tunable laser with fixed connection network (FC-FλS).
The fixed connection network consists of point-to-point
links from tunable lasers to output MUXs.
- Tunable laser with static wavelength router (WR-FλS).
The static wavelength router does not change its
configuration over time.
- Tunable laser with broadcast and select (BS-FλS). The
broadcast and select operation is time dependent and the
connection configuration can change every time-frame.
For the sake of simplicity, we do not show in figures how to
implement buffering (NIF). In principle, a tunable laser
behaves as an optical-electronic-optical conversion device,
since the incoming optical signal is converted to electronic
signal in order to modulate the tunable laser in a defined
wavelength. Thus, buffering can be embedded at the
electronic stage of tunable laser as it is done in any electronic
buffering implementation. With FλS switches using all-optical
wavelength converters rather than tunable lasers, all-optical
buffering can be implemented by parallel fiber-delay-line
approach.
A. Tunable laser with fixed connection network (FC-FλS)
1) Design and operation
Fig.3 shows the simple design of the FC-FλS for
4,2
CN
==
, which uses tunable lasers with a fixed point-to-
point connection network. DMUX separates WDM signals
into C different wavelengths. Each incoming wavelength is
fed to a tunable laser that transmits at any wavelength within
its tuning range
.
T
R The output of each tunable laser is
connected to a predefined output port. The number of fixed
connection between a pair of in-port and out-port depends on
the ratio between C and
N which defines the internal
connection ratio
.
rC N
=
A switch with
has
2r =
fixed connections between in-port/out-port pair.
Tunable lasers are tuned every time-frame, where time-
frames are derived from UTC, such that time-frames are
,
8
N =
and
16
C =

W
D
M
DMUX
W
D
M
DMUX
W
D
M
MUX
W
D
M
MUX
1 λ
2 λ
3 λ
4 λ
1 λ
2 λ
3 λ
4 λ
TL(*,1)
TL(*,1)
TL(*,2)
TL(*,3)
TL(*,4)
TL(*,2)
TL(*,3)
TL(*,4)

Fig. 3. Illustration of a 2×2 FC-FλS switch (tunable lasers are coordinated by
UTC signal, which is not shown in the figure).
switched from in-ports to out-ports without conflicts at any
out-port. Due to the nature of fixed connection system, the
color of a time-frame after being switched defines the out-
port, and hence, it defines the route it must go on.
2) Hardware complexity and scheduling feasibility
The hardware complexity of this design is CN tunable
lasers. Each input requires C tunable lasers, corresponding to
C channels. The DMUX and MUX devices are not counted
in the hardware complexity since they are identical for all the
designs described in this paper.
Scheduling time-frames using FC-FλS is rigid due to the
nature of fixed point-to-point internal connection network. To
route a time-frame along a predefined route path between
source and destination ( , )
s d , a tunable laser that receives a
signal must tune the output to one wavelength among . r For
simplicity, it is assumed that lasers have full tunable range,
that is
T
RC
=
. With this assumption, the scheduling
flexibilities of this design are given in (1) for IF, and (2) for
NIF.
C
Skrk
N


==



Proof sketch of (1): At the 1st hop, to forward a time-frame
to the 2nd hop of that predefined route, a time-frame must be
carried on 1 of r wavelengths or channels in which each
channel has k different time-frames. Hence, there are kr
scheduling choices for the 1st hop†. The following (
hops are all identical and there are only r possible schedules
at each hop. Scheduling at all hops is independent. Therefore,
the number of possible schedules is given by product of all the
()( )( )
12
...
h
krrr
is the contribution of
h hop to the combinatorial result.
Proof sketch of (2): 1st hop-based component is equal to that
of (1). For 2nd hop-based component, there are more options to
forward a TF thanks to NIF. A TF can be switched
immediately or buffered for
before being switched. Thus, for all hops except 1st one, there
are rb options to schedule a TF. The final result is given by
the product ()()
12
...krrb
3) Robustness and practical issues
Though FC-FλS has a simple design with low cost and low
control overhead, a network deployed with FC-FλSs is subject
to some disadvantages. First, it is hard to deploy different
routing protocols since routing is rigid due to the nature of
fixed internal connection network. In other words, it is
impossible to separately account for the routing and
(
FC
)
h
IFh








==

(1)
(
FC
)11
h
NIFhhh
C
N
Skr bkb
−−

(2)
1)
h−

stndth
× × ×
possible single hop schedules. ( ) .
th
h
th
bk TF durations (1
bkb
≤≤
),
()
stndth
h
rb
×× ×
.

† Note that the definition of IF is actually meaningful from the second FλS-
enabled switch only. We are analyzing FλP setup, so that the time-frame in
the first hop can be chosen freely and will represent the IF constraint for
subsequent hops.

Page 4

wavelength assignment problem if FC-FλSs are deployed,
since the wavelength assignment in one switch will determine
the route in the next. Second, for IF scheme shown in (1), the
scheduling flexibility of this design strongly depends on the
connection ratio r .
B. Tunable laser with static wavelength router (WR-FλS)
1) Design and operation
A design using tunable lasers and wavelength router (WR)
is depicted in Fig.4. The idea for this design builds on an OBS
switch design in [10]. The key characteristic of this design is
that different in-ports use different sets of channels, whose
size is r and depends on the permutation pattern, to reach the
same out-port. More specifically, in order to switch a time-
frame received by
( , )TL i j to out-port m,
to one channel among r channels defined by the designed
permutation pattern so that the transmitted time-frame can
reach
( , )MUX i m . Two common types for the selection of
fixed permutation pattern are contiguous wavelength selection
and randomized wavelength selection [10].
2) Hardware complexity and scheduling feasibility
WR-FλS requires CN tunable lasers, N modules of C C
static WRs. The scheduling feasibility of WR-FλS for both IF
and NIF schemes are given in (3) and (4).
C
SkCrk
N
( , )TL i j must tune
×

()1
1
h
h
IFh
WR
h
C
N
kN
−
−






===

(3)
()111
()
h
NIF
WR
S
hhhh
C
N
kC rbkr bNkbN
−−−






===

(4)
W
D
M
DMUX
C
λ
1λ
CN
j λ
TL(*,j)
TL(*,1)
TL(*,C)
to
out-port 1
to
out-port m
to
out-port N
fixed permutation pattern
UTC
r = C/N
inputs per each MUX
MUX(*,1)
MUX(*,N)
MUX(*,m)
Static
WR
(A) Design per in-port card
(B) An example of 2x2 switch,
4 channels per port
2
λ
1λ
MUX
MUX
W
D
M
MUX
4x4
Static
WR
TL(1,1)
TL(1,2)
4
λ
3
λ
TL(1,3)
TL(1,4)
W
D
M
DMUX
2
λ
1λ
MUX
MUX
4x4
Static
WR
TL(2,1)
TL(2,2)
4
λ
3
λ
TL(2,3)
TL(2,4)
W
D
M
DMUX
W
D
M
MUX

Fig. 4. (A) A design of WR-FλS, (B) a 2x2 switch (UTC signal is not shown).
Proof sketch of (3) and (4): The proof can be done
following the same scheme used to prove (1) and (2) with the
following modification made for the 1st hop. Using WR-FλS,
there are always kC options to select a TF for the 1st hop,
since no constraint on routing exists. For the 2nd to
an incoming TF has only r options to reach a desired out-
port, again assuming
T
RC
=
. Therefore, the products of all
hop-based components
()( ) ( )
12
...
h
kCrr
IF and NIF, respectively.
3) Robustness and practical issues
WR-FλS is another simple design. Networks using WR-
FλS has no constraint on routing since time-frame coming to
an in-port can reach any out-port. WR-FλS has no internal
conflict due to the switching nature of WR devices. However,
the scheduling feasibility is still limited by the factor . r
C. Tunable laser with broadcast and select (BS-FλS)
1) Design and operation
The illustration for per-in-port and per-out-port card design
of BS-FλS is shown in Fig.5. This design uses one tunable
laser and one broadcast-and-select
component per channel. A BSS is fabricated by the
combination of 1-to-N star-coupler (SC) and N simple
ON/OFF switching elements.
( , )
TL i j of the in-port i receives the signal of
can transmit using any channel in its tunable range. The
transmitted signal from a laser is broadcasted to all out-ports
using the star-coupler
( , )
SC i j and it is allowed to reach a
single out-port if and only if a corresponding ON/OFF
element to that port is ON. The BSS design also enables
th
h hops,
are
×
given
...
× ×
as
stnd th
× × ×
and ()()()
12
st ndth
h
kCrbrb
for
switching (BSS)
j λ and then
W
D
M
DMUX
1λ
C
λ
C
SC(*,1)
N
to out-port 1
to out-port N
SC(*,C)
N
to out-port 1
to out-port N
per in-port
TL(*,1)
TL(*,C)
UTC
UTC
ON/OFF
ON/OFF
ON/OFF
ON/OFF
BSS 1
BSS C
W
D
M
MUX
MUX(*,C)
N
N
from in-port N
from in-port N
from in-port 1
from in-port 1
C
per out-port
MUX(*,1)
a) Design per in-port card
b) Design per out-port card
UTC
UTC

Fig. 5. BS-FλS architecture.

Page 5

multicasting. All tunable lasers and ON/OFF elements are
controlled and coordinated using the UTC signal. Each time-
frame, a WDM-MUX allows a maximum C different
channels to be multiplexed to the fiber.
A BS-FλS design allows a tunable laser to transmit time-
frames to all out-ports. Moreover, BS-FλS has the advantage
over WR-FλS that a tunable laser can transmit time-frames to
any out-port using the full channel ranges
only allows to use the small fixed set of channels . r Thus,
compared to WR-FλS, BS-FλS has a larger scheduling
feasibility.

2) Hardware complexity and scheduling feasibility

The hardware requirements for BS-FλS design are: CN
tunable lasers, CN star-coupler modules,
ON/OFF devices.
The scheduling feasibility of BS-FλS design for both IF and
NIF schemes are given in (5) and (6):


==


,
C while WR-FλS
2
CN programmable
(
BS
)
h
IFhh
C
N
S kCkN




(5)
()
1
(
BS
)1
h
h
NIFhh
C
N
SkC CbkbN
−
−






==

(6)

Proof sketch of (5) and (6): For the 1st hop, there are kC
options to schedule one TF, since every channel can later be
routed following a predefined route. For the 2nd to
tunable laser can exploit all the C channels to transmit the
signal. In fact, if available TFs are found at both incoming and
outgoing channels, there is a path to schedule the
transmission. Therefore, the product of all hop-based
components for IF scheme is (
for NIF schemes is ()(
1
kC
()IF
BS
S
and
BS
S
are independent from . r The right most
expression is included only for comparison purposes with the
other architectures.
In term of scheduling feasibility, the BS-FλS design gains
h
N times compared to the WR-FλS design in both IF and
NIF schemes.
Lemma 1: If using a single SC per in-port, then the
utilization of the BS-FλS design reduces C times.
Proof sketch: Let us assume that all channels of an in-port
share a single SC. Since SC is a broadcast device, meaning
that a signal at a given input will broadcast to all outputs. At
every time-frame, strictly one and only one signal can be fed
to one of the inputs of SC, otherwise there is conflict. Hence,
if all C tunable lasers of an in-port share the same SC, at
every time-frame, only one of them is allowed to transmit,
therefore, resulting in the reduction of the utilization of the
design by C , compared to the design that deploys a single SC
per tunable laser.

Lemma 2: If the ON/OFF element is not used, then a
tunable filter can be used and scheduling feasibility is
bounded:
th
h hops, a
)
)
( )
C
...
× ×
( )
C
. Note that
12
(
...
)
h
stndth
h
kC
×× ×
, and
2
stndth
CbCb
×
()NIF
()
1
(
Filter
)
'
h
h
IFh
C
N
kC CSkN
−






≤≤
and

()
1
1(
Filter
)1
'
h
h
h NIFhh
C
N
kC CbSkbN
−
−−






≤≤

where
()
'10
CCN
=−−≥
.
TL(i,j)
j λ
TF(m,j)
SC
' j λ
in-port i
out-port m
' j λ
TL(i',j)
j λ
TF(m',j)
SC
' j λ
in-port i'
out-port m'
' j λ
' j λ
' j λ
conflict!!!
CTR

Fig. 6. One tunable filter replacing N ON/OFF switching elements.
Proof sketch: Assume that ON/OFF switches are removed
and outputs of SC devices are connected to a tunable filter
(TF), as shown in Fig.6, that is coordinately controlled by UTC
signal. At a given time-frame t ,
transmit to out-port m and
TL i j is scheduled to transmit
( , )
TL i j is scheduled to
( ', )
to out-port
'
m , both using
' j λ . Consequently, there are
( , )
TF m j and
conflicts at both inputs of
Therefore, a given tunable laser can not freely selects channel
to transmit to a given out-port, but has to watch out for the
other tunable lasers that are connected to the same filter. There
are N tunable lasers that share the same filter. Therefore, in
the worst case, a given tunable laser has only
channel options, since the other (
the other tunable
()()()
12
' ...'
h
kCCC

()()()
12
'...'
h
kCC bC b
for the NIF scheme.

3) Robustness and practical issues
BS-FλS is equivalent to a non-blocking crossbar switching
fabric since it does not introduce internal blocking. A BS-FλS
design also allows deploying multicast and broadcasting
easily.
( ', )
TF m j .
()
'1
CCN
=−−

) 1
N −
lasers.
for
channels are used by
This
the IF
yields
scheme,
stndth
×× ×
stndth
×× ×
IV. DISCUSSION
Comparisons between designs are summarized in TABLE 1.
Parameters to be compared include hardware complexity,
scheduling feasibility and routing adaptability. Design
components that are the same in all switch designs, such as
WDM-MUX and WDM-DMUX are not shown in this
comparison table.
TL
N ,
WR
N
,
N
of TLs, C C
×
static WRs, 1-to-N SCs, ON/OFF switching
elements, respectively.
SC
,
OO
N
stand for the number
To highlight the scheduling feasibility measure, we plot
some graphs of
S
and
S
/
r C N
=
(Fig.7), the hop number h (Fig.8) the buffer size b
(Fig.9).
() IF() NIF
versus the connection ratio