Download Performance Tradeoffs of Shared Limited Range Networks

Performance Tradeoffs of Shared Limited Range
Wavelength Conversion Schemes in Optical WDM
Networks
Fahad A. Al-Zahrani, Abdulgader A. Habiballa, Ayman G. Fayoumi, Anura P. Jayasumana
Electrical and Computer Engineering Department
Colorado State University, Ft. Collins, CO 80523
(Fahad, Habiball, Ayman)@engr.colostate.edu, [email protected]
Abstract— Performance of all-optical switches that employee different types of limited-range wavelength converters (LRWC) are investigated. Previous work has shown
that there is a remarkable improvement in blocking
probability while using LRWC over full range conversion,
but has not considered the coincident effect of conversion
resources sharing. We consider the case where an incoming
wavelength can be converted to a range of outgoing wavelengths, where d is the range of conversion. The simulation
results demonstrate that the performance improvement
obtained by full range wavelength conversion can almost
be achieved by using a fractional ranged ranged LRWC.
I. I NTRODUCTION
The optical switches used to aggregate data through
the optical backbone must exhibit increased efficiency
while maintaining low cost. Wavelength Converter (WC)
[4], which is used to manipulate optical channels in
Dense Wavelength Division Multiplexing (DWDM), optical switches, routers, and add/drop multiplexers, represent a very important resource that is essential for
exploiting the capacity of the optical network. WC is
used to eliminate the severity of ”wavelength continuity
constraint” [10], and improving wavelength agility for
dynamic reallocation of optical channels, which in turn
enhances the scalability of the network to match the
demand for bandwidth. The proper use of converters
decreases the number of necessary wavelengths in an
optical network. Moreover, tuneability of a WC further
enhances utilization and flexibility. Since the number
of available wavelengths is a limited resource, and it
also affects the complexity of the switches, different
WC ranging architectural configurations must be studied
and analyzed to achieve an adequate level of wavelength
utilization while maintaining the efficiency.
Conventional optoelectronic conversion, which uses
Optical/Electrical/Optical (OEO) line cards that incorpo0-7803-9019-9/05/$20.00 ©2005 IEEE.
rates a tunable laser for its output, incurs a high cost due
to the complex implementation of such converters. Optical networks with limited range wavelength conversion,
LRWC, will be less costly to implement than networks
that have no constraints on conversion [8], [12]. The
conversion within each node could be accomplished alloptically instead of using OEO technique. All-optical
WC [9] is of increasing interest for future all-optical
telecommunication networks. In these all-optical devices
[13], the conversion efficiency is a function of the input
and output wavelengths, leading to limited conversion
capabilities. Even otherwise, the number of such shared
devices in the node can be reduced while maintaining
comparable efficiency requirement. However, it is difficult to provide a full capability of WC without OEO
conversions under the current technology.
Converters sharing, [5], [1], [2], has led the way for
new generation of wavelength routers, called wavelength
convertible switches or WCSs. The different types of
WCS [14] architectures have introduced different WC
sharing schemes in order to minimize the cost and
maintain efficiency. It has been shown that these WCSs
with shared conversion resources can achieve very close
performance to full WC when every channel has its own
fully ranged dedicated converter [1], [2].
Previous work has shown that LRWC can considerably improve the performance of all-optical networks,
[8], [12]. However, the concept of conversion resource
sharing in conjunction with different wavelength range
configurations has not been explored. In this paper, we
perform a simulation based analysis of different LRWC
configurations that implement conversion resources sharing. This paper focuses on minimizing the cost associated with converters and maintaining an appropriate
level of performance by quantifying the penalty for
mismanaging WC resources.
This paper is structured as follows. Section 2 presents
LRWC. Sections 3 presents different LRWC ranging
architectures. Simulation results are discussed in Section
4 followed by conclusion in Section 5.
II. WAVELENGTH C ONVERSION
The optical layer transports a large numbers of channels, each operating at rates such as 2.5 Gb/sec or
10 Gb/sec, between two nodes using optical DWDM
technology [4], [8], which requires WC mechanisms to
resolve conflicts. Ideally, the conversion process should
be independent of bit rate or signal format, entirely
optical, require little power, not degrade the signal, and
be tunable. All-optical WC promises better response,
simplicity, and transparency to the signal format whereas
optoelectronic converters are inherently dependent on
input transmission format and bit rate, so a change in
data rate requires changing the converter, [4].
A. All-optical Wavelength Converters
All-optical networks will eliminate the electronic bottleneck and deliver scalability, and flexibility. Related
cost constraints mean that WCs must be compact, and
low-cost relative to existing optoelectronic WC solutions.
Low cost and small size make the semiconductor optical
amplifier an attractive option. The nonlinear phenomena
in semiconductor optical amplifiers have been widely
utilized for WC and optical switching as follows:
• A semiconductor optical amplifier, or SOA [11], can
be used to improve a signal that has weakened. It
outputs a new wavelength that has same modulation
as input wavelength. The wavelength can be engineered to vary within a very large range between
1300 and 1600 nm.
• Four Wave Mixing, [6], is made possible by nonlinear effects between two closely spaced wavelengths, that produce conjugate outputs from both
initial wavelengths. Cross-phase modulation converters use the input signal to modulate the phase
of a second wavelength. This effect can be used to
convert an input signal at a specific wavelength to
an exact copy of the signal at a different wavelength.
• Good results have been obtained with parametric conversion in periodically poled lithium-niobate
waveguides [3]. It is a very fast conversion method
relying on interaction between a signal and highpower pump beam inside the nonlinear crystal.
Single or multi-channel WC is achieved in the
optical chip. The WC chip is protocol independent
and currently supports 10 Gbit/s, 40 Gbit/s, and the
future extension to 160 Gbit/s high-speed network
applications. The operation of this type covers the
entire C-band.
B. Limited Range Wavelength Conversion, LRWC
LRWC means that each input wavelength may be
converted to any wavelength of a specific range of wavelengths. All-optical LRWC decreases blocking probability, increases NxN switch scalability, lowers power consumption, and operates at speeds important for telecommunication. Even though LRWC does not eliminate the
severity of wavelength continuity problem completely, it
enhances the performance to levels close to full range
WC and reduces the cost in term of implementation and
control. The cost of conversion not only depends on the
number of converters in the network, but also considers
the architectural complexity of each individual converter.
A significant cost penalty can be incurred if WC is not
maintained in an efficient manner. However, we show
how one could significantly mitigate this penalty by
allowing limited range WC, using a small subset of
wavelengths as the range, complemented with the ability
to share these resources.
III. WC R ANGING A RCHITECTURES
This study examines how WC is related to efficient
utilization of an optical channel. A recent advancement
in all-optical WC technology makes it possible to convert
the input wavelength to different output wavelength if
those are within different limited ranges. Implementing
all-optical full WC is quite difficult due to technological
limitations. The transmitters and receivers for all-optical
wavelength converters are tunable [7]. The tuning range
is, perhaps, determined at the time of fabrication and
can not be changed thereafter. Presumably, range tunable
transmitters and receivers are cheaper and simpler to
construct than fully ranged ones. Our aim is to determine
the range configuration that is most practical for tunable
all-optical devices in the presence of WC resources
sharing.
In [8], [12], the authors have analyzed all-optical
networks without consideration for WC sharing or the
architectural differences in defining the conversion range.
Notably, our study allows the observation of the interactive behavior of LRWC with conversion sharing and
the tradeoff scenarios that could result in a cheaper
conversion cost.
We defined a measure of WC called the conversion
range. If a network node has a conversion range d then a
wavelength may shift to any one of a pre-defined set of d
A. Simulation of WC Ranging Schemes
Networks with different conversion ranging architectural definitions and different conversion sharing degrees were simulated considering c, number of shared
converters in the conversion bank, W, total number of
wavelengths available on a fiber, and d. If the requested
wavelength is not available in any outgoing fiber, the
request will be forwarded to one of the shared converters
within the pre-specified range d for the given architecture.
The following assumptions are used in our simulation
models. 1) The offered traffic is in the form of connection requests for an entire optical channel. These
requests arrive at each node according to an independent
stationary Poisson process with rate λ. 2) Connection
holding time is exponentially distributed with the mean
1/µ. 3) Connection request that cannot be serviced are
blocked. 4) Wavelengths are assigned randomly from the
set of free wavelengths on the associated path. Existing
requests cannot be reassigned different wavelengths. If
the input/output range of the converter is defined as Any-
to-Any, the request will be forwarded to any one of the
shared output converters, where d=W for the Any-to-Any
case. In this case, if there is no free wavelength available
on the outgoing range, or there is no free converter, then
the request is blocked.
Both Any-to-Any and Any-to-Range conversion architectures, which are easily implemented as all-optical
devices, have the same conversion bank behavior, but
they differ in the available output wavelength range that
can serve the path setting request. The Any-to-Range
scheme is expected to result in a blocking probability
performance comparable to the earlier scheme, where
this performance can be further enhanced by adjusting or
managing the conversion resources sharing degree. In the
Range-to-Any and Range-to-Range LRWC simulations,
the behavior of the conversion bank differs from the
previous two cases due to the fact that each request
uses just one specific converter that has the incoming
wavelength in its range.
IV. S IMULATION R ESULTS
In this section, we discuss different simulation results
for different LRWC ranging schemes. In figure 1, the
blocking is plotted as a function of offered load per
wavelength for 4 differently ranged LRWC architectures. We considered 3 different Any-to-Any switch sizes,
W=64, 32, 16, that have similar shared conversion degree
of d=2, where d=W/c. The results for this architecture
consistently show that switches with higher number of
wavelengths, W, have better blocking performance given
similar d. The graph also shows that at lower offered
0.45
0.4
0.35
Blocking probability
wavelengths. The cost of all-optical devices, packaged in
a way that both transmitter and receiver are in the same
chip, is a function of the complexity in ranging both
the transmitters and receivers. The ranging architectures
proposed and simulated in this paper have LRWC in the
following sense.
• Any-to-Any: The conversion device based on this
architecture can not fall under all-optical category
due to technology limitations. This device can convert any incoming input wavelength to any outgoing
one, corresponding to full range conversion. Even
in optoelectronic converters, wider range tunable
transmitters are more difficult to build and accordingly more expensive.
• Any-to-Range: This conversion device can convert
any incoming wavelength to a wavelength that falls
in a pre-specified range for full non overlapping
coverage of wavelengths, which is considered in
this paper, d=W/c, where W is the total number of
wavelengths in the fiber and c is the number of
converters shared in the node.
• Range-to-Any: This device behaves in an opposite manner compared to the previous ones. The
incoming request is required to use one specific
conversion device depending on its wavelength and
the range it falls into.
• Range-to-Range: The range in both sides are defined
by d=W/c.
0.3
0.25
Any−to−Any, W=16, C=8
Any−to−Range, W=16, C=8
Range−to−Any, W=16, C=8
Range−to−Range, W=16, C=8
Any−to−Any, W=32, C=16
Any−to−Range, W=32, C=16
Range−to−Any, W=32, C=16
Range−to−Range, W=32, C=16
Any−to−Any, W=64, C=32
Any−to−Range, W=64, C=32
Range−to−Any, W=64, C=32
Range−to−Range, W=64, C=32
0.2
0.15
0.1
0.05
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Offered load per wavlength
Fig. 1.
Blocking Vs. Offered Load per Wavelength
1
0.5
Any−to−Range, C=32, d=2
Any−to−Any, C=2
Any−to−Any, C=4
Any−to−Any, C=8
Any−to−Any, C=16
Any−to−Any, C=30
Any−to−Any, C=32
Any−to−Range, C=30, d=2
0.45
0.4
0.35
Blocking Probability
loads, all 3 cases have similar performances due to the
abundance of wavelengths and conversion resources with
respect to load serviced. In the same graph, we also
considered 3 different Any-to-Range switch sizes, W=64,
32, 16, that have similar shared conversion degree, d,
to the preceding case. The results showed that Any-toRange architecture’s blocking performance behaves in a
comparable manner to the Any-to-Any case implying that
the front ranging mechanism does not degrade performance while minimizing the architectural complexity of
the switch. This behavior, also, creates an opportunity for
network cost, that depends on number of converters, and
conversion cost, that depends on converter architecture,
tradeoffs.
In a similar fashion, we considered switch sizes
of W=64, 32, 16 and d=2 for Range-to-Any and
Range-to-Range cases. These two architectures’ blocking
performances are considerably worse than Any-to-Any
and Any-to-Range due to the one converter limitation
imposed, resulting from the end ranging mechanism, on
all incoming requests. This limitation makes the switch
behave the same regardless of the size, W, implying that
the end ranging scheme can severely degrade performance. The Range-to-Range scheme is the least practical
due to the fact that its behavior is severely affected by
the ranging limitation imposed on both the input and the
output sides.
From pervious results, we can expect that Rangeto-Any and Range-to-Range architectures to be less
sensitive to the change in number of converters, and
accordingly to d, due to the already imposed ranging
limitation on their input sides. This raises a question
about how sensitive the Any-to-Any and Any-to-Range
schemes to the change in number of converters c and
the conversion range d. In figure 2, for an LRWC with
W=64 and c=32, 30, 16, 8, 4, and 2, the blocking of Anyto-Any and Any-to-Range architectures decreases with
converter count, even though, the width of the conversion
range, d, gets smaller. The graph also shows that both
Any-to-Any and Any-to-Range architectures behave comparably in each scenario indicating that Any-to-Range
ranging scheme is a valid architectural abstraction that
provides an equivalent performance behavior to Any-toAny scheme. This reaffirms the preceding results and
show that Any-to-Range is the most practical and the
most comparable to the Any-to-Any architecture.
According to the relationship between the conversion
range d and the number of shared converters, the range
of each converter is expected to be less with increased
sharing. Accordingly, the Range-to-Range architecture
0.3
0.25
0.2
0.15
0.1
0.05
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Offered load per wavelength
Fig. 2.
Blocking Vs. Offered Load per Wavelength
suffers the most due to ranging restriction imposed
on both sides when the number of shared converters
increases resulting in performance degradation. It also
can be deduced that these conversion ranging schemes
are more responsive to the change in number of shared
converters than the change in the conversion range, even
though, both are interrelated according to d=W/c.
Finally, the offered load versus the gain, which is defined as the ratio of utilization of the Any-to-any architecture versus that of the architecture of interest, shown in
figure 3, for W=64, c=32 and d=2, confirms that the Anyto-Range architecture performs in an analogous fashion
to the fully ranged Any-to-Any method. It also shows
that Range-to-Any architecture lags in performance due
to the input limitation imposed on the incoming traffic
where as the Range-to-Range method is the poorest due
the dual limitation imposed on both sides of the switch
as explained earlier. These performance differences are
clearer at higher offered loads due to the facts that
limited range architectures are more susceptive to lag in
utilization performance with higher traffic demand. The
differences in these architectures define the practicality
of each case with respect to a range of offered loads.
This indicates that more limiting, conversion range wise,
architectures can be employed in lightly loaded networks
increasing the cost efficiency on both the WC hardware
and control architecture of the switch. It should be noted
that ranging of LRWC architectures can be implemented
in an overlapping fashion causing a relative improvement
in performance at the expense of control complexity.
R EFERENCES
1
Any−to−Range with respect to Any−to−Any
Range−to−Any with respect to Any−to−Any
Range−to−Range with respect to Any−to−Any
0.95
Gain
0.9
0.85
0.8
0.75
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Offered load per wavelength
Fig. 3.
Gain Vs. Offered Load per Wavelength
V. C ONCLUSION
This paper investigated issues associated with LRWC
in the presence of conversion resources sharing. It analyzes the impact of using different architectures for
wavelength conversion to mitigate the cost penalty. The
throughput improvement due to the use of LRWC, in
the presence of conversion sharing, depends more on the
degree of sharing than the conversion range. The study
shows that smaller conversion ranges can approach the
same performance as full-range WC at slightly higher
degree of sharing. The implementation of Any-to-Range
architecture in all-optical wavelength converters would
be more efficient and elegant solution, allowing fully
integrated optical cross-connect/wavelength conversion
boxes. Any-to-Range means less complex and less costly
hardware than full range conversion, yet results in acceptable channel utilization. This introduces the issue of
cost effectiveness when considering different switching
architectures, conversion options, and hop configurations.
The limited ranged architecture reduces the need
for extremely complex ”any-to-any” control algorithms.
Nevertheless, these ranging architectures do give up
some flexibility and overall network efficiency. The
complexity and the cost of converters depend on its
conversion ranging mechanism. Switches with limited
range conversion will be less costly to implement than
the ones that employ full ranged converters. The cost
of implementing optical layer can be expected to scale
roughly in proportion to the number of conversion resources and how the resources are architected.
[1] F. Al-Zahrani, A. Habiballa, and A. Jayasumana, Path Blocking
Performance in Multi-Fiber Wavelength Routing Networks with
and without Wavelength Conversion, Proc. 12th International
Conference on Computer Communications and Networks, 2003,
580-583.
[2] F. Al-Zahrani, A. Habiballa, and A. Jayasumana, Performance
Merits of Multi-Fiber DWDM Networks Employing Different
Shared Wavelength Conversion Resources Architectures, Proc.
IEEE region 5 conference, 2004, 59-67.
[3] D. Caccioli, A. Paoletti, et al, Field Demonstration of in-Line
All-Optical Wavelength Conversion in a WDM Dispersion Managed 40-Gbit/s Link, Selected Topics in Quantum Electronics,
IEEE Journal of, 10 (2) , 2004, 356-362.
[4] J. M. H. Elmirghani and H. T. Mouftah, All-Optical Wavelength Conversion: Technologies and Applications in DWDM
Networks, IEEE Commun. Mag., Mar ,38 (3), 2000, 86-92.
[5] A. Habiballa, F. Al-Zahrani, A. Jayasumana, Wavelength Conversion Resources Allocation Algorithms for Share-per-Link
Wavelength Convertible Switch, Proc. IEEE region 5 conference, 2004, 131-139.
[6] K. Inoue, T. Hasegawa, K. Oda, and H. Toba, Multichannel Frequency Conversion Experiment Using Fiber Four-Wave Mixing,
Electron. Lett., 29 (19), 1993, 1708-1710.
[7] T. Lin and G. Sasaki, Nonblocking WDM Networks with
Fixed-Tuned Transmitters and Tunable Receivers, Proc. 37th
Annual Allerton Conference on Communication, Control, and
Computing, 1999, 400-401.
[8] R. Ramaswami and G. Sasaki, Multi-Wavelength Optical Networks with Limited Wavelength Conversion, IEEE/ACM Trans.
Networking, 6 (6), 1998, 744-754.
[9] L.H. Spiekman, U. Koren, M.D. Chien, B.I. Miller, J.M.
Wiesenfeld, J.S. Perino, All-Optical Mach-Zehnder, Wavelength
Converter with Monolithically Integrated DFB Probe Source,
IEEE Photonics Technology Letters, 9 (10), 1997, 1349-1351.
[10] T. Tripathi and K. N. Sivarajan, Computing Approximate Blocking Probabilities in Wavelength Routed All-optical Networks
with Limited Range Wavelength Conversion, IEEE J. Select.
Areas Commun., 18 (10), 2000, 2123-2129.
[11] D. Wolfson, P. B. Hansen, T. Fjelde, A. Kloch, C. Janz, A.
Coquelin, I. Guillemot, F. Gaborit, F. Poingt, and M. Renaud,
40Gbit/s All-Optical Wavelength Conversion in an SOA-Based
Allactive Mach-Zehnder Interferometer, ECOC’99, 1999, II170-171.
[12] G. Xiao and Y. W. Leung, Algorithms for Allocating Wavelength Converters in All-optical Networks, IEEE/ACM Trans.
Networking, 7 (4), 1999, 545-557.
[13] L. Xu, L.K. Oxenlowe, N. Chi, J. Mork, P. Jeppesen, K. Hoppe,
J. Hanberg, Basic Characterization of Wavelength Conversion
at 40 Gb/s Based on Electroabsorption Modulators, Proc. IEEE
Lasers and Electro-optics Society 2002 Annual meeting, 2002,
111-112.
[14] J. M. Yates and M. P. Rumsewicz, Wavelength Converters in
Dynamically Reconfigurable WDM Networks, IEEE Communications Surveys, Second quarter, 1999, 2-15.