Content uploaded by Pouria Sayyad Khodashenas
Author content
All content in this area was uploaded by Pouria Sayyad Khodashenas on Oct 16, 2014
Content may be subject to copyright.
Khodashenas et al. VOL. 5, NO. 6/JUNE 2013/J. OPT. COMMUN. NETW
1
Abstract—Elastic optical network (EON)
technology arises as a promising solution for future
high-speed optical transport, since it can provide
superior flexibility and scalability in the spectrum
allocation for seamlessly supporting diverse services,
while following the rapid growth of Internet traffic.
This work focuses on lightpath adaptation under
time variable traffic demands in EONs. Specifically,
we explore the possibility of utilizing the spectral
fragmentation to increase the spectrum allocation
capabilities of EONs. In this context, a heuristic
Spectrum Allocation (SA) algorithm, which
intentionally increases the spectral fragmentation in
the network is proposed and validated. In our
proposal, the spectrum assigned to each new
connection is in the middle of the largest free
spectral void over the route, aiming to provide
considerable spectral space between adjacent
connections. These free spectral spaces are then used
to allocate time-varying connections without
requiring any lightpath reallocation. The obtained
simulation results show a significant improvement in
terms of network blocking probability when utilizing
the proposed algorithm.
Index Terms—Elastic optical network; Time-
varying connections; Routing and Spectrum
Allocation.
I. INTRODUCTION
riggered by emerging services such as high-definition
video distribution or social networking, the IP traffic
volume has been exponentially increasing to date.
Furthermore, the traffic growth rate will not stop here due
to the day by day technology advances. For example, new
Manuscript received November 16, 2013; accepted …….;
published ….. (Doc. ID ……..).
The authors are with Advanced Broadband Communications
Center (CCABA), Universitat Polit`ecnica de Catalunya (UPC),
Jordi Girona 1-3, 08034 Barcelona, Spain. Tel: (+34) 93 401 7179,
Fax: (+34) 93 401 7200 (email: pkhodashenas@tsc.upc.edu)
hardware advances such as multicore processing,
virtualization and network storage will support a new
generation of e-Science and grid applications requesting
data flows of 10 Gb/s up to terabit level. The predictable
consequence is that network operators will require a new
generation of optical transport networks in the near future,
so as to serve this huge and heterogeneous volume of traffic
in a cost-effective and scalable manner [1]. In response to
these large capacity and diverse traffic granularity needs of
the future Internet, the elastic optical network (EON)
architecture has been proposed [2]. By breaking the fixed-
grid spectrum allocation limit of conventional wavelength
division multiplexing (WDM) networks, EONs increase the
flexibility in the connection provisioning. To do so,
depending on the traffic volume, an appropriately sized
optical spectrum portion is allocated to a connection in
EONs. Furthermore, unlike the rigid optical channels of
conventional WDM networks, a lightpath can expand or
contract elastically to meet different bandwidth demands in
EONs [3]. In this way, incoming connection requests can be
served in a spectrum-efficient manner.
This technological advance poses additional challenges on
the networking level, specifically on the efficient connection
establishment. Similar to WDM networks, an elastic optical
connection must occupy the same spectrum portion between
its end-nodes, that is, ensuring the so called spectrum
continuity constraint. In addition, the entire bandwidth of
the connections must be contiguously allocated, also
referred as the spectrum contiguity constraint. The Routing
and Spectrum Allocation (RSA) problem in elastic optical
networks has been widely studied, putting more emphasis
on dynamic network scenarios [4-7]. There, connection
arrival and departure processes are random and the
network has to accommodate incoming traffic in real time.
Considering the near future technology advances (e.g., high
capacity bandwidth variable transponders) and the
exponential increase of network traffic, it is foreseeable to
have large intervals between the establishment and the
release of connections. Thus, during such relatively long
periods, the bit-rate demand of any connection may
fluctuate following short- and mid-term traffic variations.
Although the EON technology enables flexible adaptation of
connections to such time-varying traffic demand changes,
Using Spectrum Fragmentation to
Better Allocate Time-Varying
Connections in
Elastic Optical Networks
Pouria Sayyad Khodashenas, Jaume Comellas, Salvatore Spadaro, Jordi Perelló,
Gabriel Junyent
T
Khodashenas et al. VOL. 5, NO. 6/JUNE 2013/J. OPT. COMMUN. NETW
2
the spectrum fragmentation issue prevents high resource
utilization in EONs. In light of this, various spectrum
defragmentation proposals, requiring the periodic or on-
demand re-allocation of connections, have been introduced
[8,9]. The basic idea behind these proposals is to consolidate
fragmented spectrum bands, aiming to increase the
probability of finding sufficient contiguous spectrum to
accommodate future connection requests over the network.
Despite spectrum defragmentation schemes can improve
the network efficiency under time-varying traffic,
connection re-allocation increases the complexity and cost of
the network. Moreover, advanced optical devices like
wavelength converters and tunable optical switches are
necessary for realizing such proposals, further increasing
the complexity and also cost of the network. Finally, it is
worth mentioning that a significant amount of control plane
signaling messages are required to setup and release
lightpaths in the network, which burdens the control plane
in the network with additional overhead.
While spectral voids between adjacent connections (due
to fragmentation) are traditionally considered as a problem,
this statement does not always apply. Indeed, as long as
connections required bandwidth may grow over time, free
spectral voids are crucial to accommodate additional
bandwidth demands without requiring the re-allocation of
the already established connections (an existing connection
can easily adapt to transmission rate fluctuations if it has
free spectral voids around it). Therefore, an alternative
approach could consist in deliberately leaving space
between connections. In light of this, we propose a heuristic
Spectrum Allocation (SA) algorithm which intentionally
increases spectral fragments over the network in order to
boost time-varying connections allocation. In our proposal,
each new elastic connection is allocated in the middle of the
largest spectral void over the selected end-to-end route. As a
result, the possibility of serving short- to mid-term bit rate
fluctuations increases significantly. The rest of paper is
organized as follows. In Section II we initially review the
principles of serving time-varying connections in EONs.
Section III details the proposed algorithm. Simulation
results are presented in Section IV. Finally, Section V
concludes the paper.
II. ELASTIC OPTICAL NETWORKS AND TIME-VARYING
TRAFFIC
In this section, we initially introduce the problem of
serving time-varying connections in EONs and review some
previous contributions on this topic. Then, applicable
lightpath adaptation policies are also introduced.
A. Time-varying Traffic
To accommodate traffic in EONs, the total available
spectrum is divided into constant spectrum units with a
granularity finer than the typical 50-GHz grid used in
WDM systems (e.g., 12.5 GHz), referred as frequency slots
(FS). For example, 1, 1.5 and 2 THz spectrum correspond to
80, 120 and 160 FSs, respectively. Each FS can carry some
bit rate depending on the modulation format used [10]. For
the sake of simplicity, we will assume a modulation
efficiency of 1b/s/Hz in this work. A connection is served by
assigning a route and allocating a set of contiguous FSs on
all links along it.
Initially, the problem of static RSA received the research
community's attention. In this scenario, all the requested
connections are given, as a traffic matrix known in advance.
Then, connections are served under the constraint that no
spectrum overlapping is allowed among them. The solution
gives the route and the allocated frequency slots to every
connection, while minimizing a given performance metric,
such as the total utilized spectrum. Some studies like those
in [11-13] focus on the routing, modulation level and
spectrum allocation (RMLSA) in EONs. They formulated
the problem and proved that it is NP-complete. Integer
linear programming (ILP) formulations, as well as heuristic
algorithms for solving such a complex planning problem
were presented in these studies.
While minimizing the total required spectrum for serving
a static traffic matrix is important in the planning phase of
EONs, there is an increasing need for dynamic lightpath
provisioning, as traffic fluctuates over time in current
networks. Thus, the community's attention also focused on
the dynamic spectrum allocation problem. Authors in [4]
triggered this effort by proposing a general algorithm to
assign frequency slots in EONs under dynamic traffic.
Meanwhile, they proposed the first distance adaptive
dynamic routing and frequency slot assignment in [5],
concluding that the necessary bandwidth for serving
connections can be significantly reduced by employing a
distance adaptive proposal. In [14], the dynamic allocation
and release of connections ranging from 10 Gb/s to 400 Gb/s
were investigated. This study shows that by introducing
technologically advanced devices, such as high capacity
flexible optical transponders, the holding time of the optical
connections in EONs increases significantly. During such a
long period, the connection bit rate may vary as a function
of the time. Thus, the spectrum allocated to the connection
(lightpath) has to flexibly change, so as to adapt to the
variation in the requested bit rate. Fig. 1 presents the
spectrum utilization of an exemplary link in an EON with
time-varying connections at two different moments. As
shown in Fig.1-(a), connections over the exemplary link
carry traffic between different end nodes. A specific set of
FS (which are highlighted in the figure) is assigned to each
connection. Fig.1-(b) shows the spectrum utilization of the
Fig. 1. Spectrum allocation of an exemplary link with time-
varying connections. Two different time instances are displayed in
(a) and (b). Free frequency slots between adjacent connections are
used to accommodate time-varying traffic changes.
Khodashenas et al. VOL. 5, NO. 6/JUNE 2013/J. OPT. COMMUN. NETW
3
same link at another instant, where the bandwidth required
by connection 2 has doubled. As shown, the number of FS
allocated to this connection has changed to adapt to the
required transmission rate. It is worth to note that the free
frequency slots between adjacent connections are used to
accommodate the mentioned traffic change. There exist
different policies for such a lightpath adaptation in the
literature [15-17]. We will describe the one used in this
paper in the next subsection.
B. Lightpath adaptation policies
The way how connections adapt their spectrum to the
instantaneous required bit rates is called lightpath
adaptation policy. Authors in [15] and [16] studied for the
first time different policies through extensive simulations
results. They continued their work in [17], by introducing
some mechanisms to the previously proposed policies,
aiming to strike a balance between network blocking
probability and the number of re-allocated connections.
These policies can be summarized as follows:
Constant Spectrum Allocation (CSA): A fixed
number of FS are reserved around each connection.
No spectrum sharing is permitted among adjacent
connections. No connection re-allocation is
permitted.
Dynamic Alternate Direction (DAD): A connection
wishing to increase its bit rate alternates between
using its higher and lower FS, until it reaches an
already occupied slot. If additional slots are
needed, the symmetry of the expansion is lost but
the connection continues to expand towards the
only possible direction. No connection re-allocation
is permitted.
Avoid Close Neighbors (ACN): Each connection
expands towards the opposite direction of its closest
neighboring connection on any of the links along its
path. Keeping the same principle, contraction is
conducted from the direction of the closest
neighboring connection. No connection re-allocation
is permitted.
Shift-ACN: This is an enhanced version of ACN
that allows connection re-allocation. If there are no
available FS in the maximum allowable expansion
region, neighboring connections are shifted in a
way that maximizes the minimum available FS
among all of its neighbors. Note that connection re-
allocation is only allowed to direct neighboring
connections of the connection requesting a
spectrum expansion.
Float-ACN: This is an enhanced version of Shift-
ACN. In this version, all connections are free to
float in the spectrum as they are pushed by their
neighbors. Thus, connection re-allocation is not
only restricted to the direct neighboring
connections of the connection requesting a
spectrum expansion, as it was in the previous case.
It is worth to mention that, to avoid traffic interruption
during the extension process, hitless techniques such as the
push-pull technique in [18,19] are applied.
Considering the aforementioned cases, we propose an
enhanced version of the DAD policy, namely Shift-DAD
policy in this article. Note that, since the connections can
contract without any extra efforts, here we explain only the
extension part of this policy.
Shift-DAD policy:
1) Calculate the necessary number of FSs to satisfy
the connection’s bandwidth change. Set this
number as the required FSs.
2) Count the number of free FSs at both sides of the
connection that needs to expand. In case of it less
than required FSs go to Step 3, otherwise go to
Step 4.
3) Check the spectral status of the path to find the
first void with enough FSs to accommodate the
connection. If found, re-allocate it. Otherwise, do
not extend the connection and add the amount of
bandwidth change into the total bandwidth
dropped. Finish.
4) Check the left side of the connection for a free FS;
if it exists, extend the connection by one FS
toward the left. Decrease the required FSs by
one.
5) If the required FSs are higher than zero, check
the right side of connection for a free FS; if it
exists, extend the connection by one slot toward
the right. Decrease the required FSs by one.
6) If the required FSs are still higher than zero, go
back to Step 4. Otherwise, Finish.
Similar to the DAD policy, a connection wishing to
increase its bit rate alternates between using its higher and
lower FSs, until reaching an already occupied slot. If
additional slots are needed, the symmetry of the expansion
is lost but the connection continues to expand towards the
only possible direction. When a connection needs more FSs
than those available around it, the connection is re-
allocated to the spectrum void with enough bandwidth
along the connection’s path. Note that, if a void large
enough to re-allocate the connection is not found along the
path, the entire bandwidth expansion operation is blocked.
An alternative to this could have been to permit partial
satisfaction of the bandwidth expansion. Nonetheless, this
has been left for future study.
In practical cases that use conventional spectrum
allocation algorithms such as First Fit, it happens that
connections are established very close to each other over the
network, aiming to increase the available spectrum for new
arriving connections. With these approaches, it is hard to
find enough bandwidth around a lightpath to adapt it to its
traffic fluctuation. Meanwhile, since the occupied portion of
spectrum is compacted in one side of available spectrum
(lower side of spectrum), it is possible to have enough FS for
serving time varying connections in the upper side of
spectrum. Therefore, by re-allocating connections over the
network in a hitless fashion, a significant improvement in
time-varying traffic allocation can be achieved.
Despite its potential benefits, connection re-allocation
Khodashenas et al. VOL. 5, NO. 6/JUNE 2013/J. OPT. COMMUN. NETW
4
increases the complexity and the cost of network. As it was
stated before, many advanced photonic devices are needed,
as well as a large amount of control plane overhead. In light
of this, it is important to make a good trade-off between the
successfulness in allocating time-varying connections and
network complexity and cost-efficiency.
Moreover, it is worth to highlight that SA algorithms
have a great impact on the efficacy of serving time-varying
connections in EONs. Indeed, SA algorithms can achieve
good arrangement of the connections in the space and
frequency domains, thus increasing the possibility of
serving and expanding time-varying connections in the
network. An appropriate SA algorithm serves the
connection requests, by assigning FSs in a way that
minimizes the average network blocking probability, taking
into account the requirements of the lightpath adaptation
policy. To achieve the previously mentioned goals, next
section details a novel SA algorithm called Mid Fit.
III. SPECTRUM ASSIGNMENT ALGORITHM
There exist various Spectrum Allocation (SA) approaches
to allocate incoming connections with different bandwidth
granularities in EONs [20]:
First Fit: The connection request is placed in the first
spectral gap along its route fitting the requested
bandwidth. This algorithm is used for benchmarking
purposes in this work.
Smallest Fit: The connection request is placed in the
smallest available spectral band along its route.
Random Fit: Any available spectrum portion along the
route, with enough space to allocate the connection
request, can be randomly selected to allocate it.
In contrast, in this paper we propose a novel heuristic
called Mid Fit SA approach. Its main objective is to
maximize the spectral voids between adjacent connections,
thus increasing the possibility to successfully serve their
potential bandwidth increments. To achieve this goal, after
calculating the candidate path between source and
destination nodes of the connection request, this one is
established in the middle of the largest contiguous
spectrum void along the calculated path. Specifically, if
multiple same-sized spectral bands exist, the one placed in
the lowest part of the spectrum is selected. In this way, the
spectral distance between adjacent connections is
maximized. The available spectrum left between
connections can later on be used to serve connection
bandwidth fluctuations. It is worth to mention that
lightpath bit rate adaptation can then be realized without
any spectrum conversion or re-allocation. This implies a
significant reduction in the network cost, complexity and
energy consumption. In fact, bit rate fluctuations are
allocated by just increasing the bandwidth (number of FSs)
assigned to the connection (assuming that the total
connection bit rate is lower than the whole capacity of the
transmitter).
Fig. 2 illustrates a simple 4-node network, whose current
spectral status is shown in Fig.2-(a). There exist some
connections with different path lengths (a 2 FSs connection
from D to B and a 1 FS connection from B to C). If a new
connection request arrives, for example a 1 FS connection
between nodes A and C, the candidate path for establishing
the connection is firstly calculated. Let us assume that this
path consists of links AB and BC. The spectral status of this
candidate path is shown in Fig.2-(b). As illustrated, there
exist two spectral voids in the spectrum of the candidate
path, G1 and G2. According to our proposal, the Mid Fit
algorithm assigns the central FS of G2 to the connection, as
shown in Fig.2-(c). The free frequency slots between
existing connections are used to serve data rate
fluctuations. By applying this spectral allocation strategy
the bit rate of the connection can be doubled and even
triplicated in the future without any need for connection re-
allocation. That is, bit rate fluctuations are accommodated
in the network by only increasing the bit rate of
transmitters (assuming that the final bit rate is less than
the whole capacity of the transmitter).
In contrast, let us consider a conventional SA algorithm
such as First Fit for spectrum allocation in this example.
After calculating the candidate path, the first gap with
Fig. 3. NSFnet network topology: 14 nodes and 21 fiber links.
Fig. 2. Illustrative example: (a) Simple 4-node network; (b)
spectral status of the candidate path; (c) FS status using the Mid
Fit SA approach; (d) FS status using conventional First Fit SA.
Khodashenas et al. VOL. 5, NO. 6/JUNE 2013/J. OPT. COMMUN. NETW
5
enough FSs (G1) is selected for establishing the new
connection request, as shown in Fig.2-(d). Hence, since
there is no space between adjacent connections, there is no
chance for serving the connection data rate fluctuations
without re-allocating it. Indeed, it could be possible to
accommodate the connection in the network upon expansion
(i.e., over spectral void G2). However, re-allocation would
have to be unavoidably employed in this case.
IV. SIMULATION RESULTS
We evaluate the performance of the proposed Mid Fit SA
approach through extensive discrete event simulation
studies, and compare its performance with a First Fit SA
[10] in a scenario with time-varying elastic connections. In
both cases, a k-Shortest Path routing strategy, with k=3 is
used. During the simulations, existing connections are
allowed to change their bit rate, so that the spectrum they
use can be dynamically expanded and contracted
accordingly. Time-varying connection selection starts after
an initial transient phase, in which, 104 connection requests
are simulated (i.e., enough to achieve steady state network
operation). Once this initial transient phase ends, a total
number of 5x106 connection requests are simulated, which
allows us to get statistically relevant Blocking Probability
(BP) results. Furthermore, a percentage of the established
connections are randomly selected and triggered to change
their bandwidth. This percentage of time-varying
connections with respect to the total of established
connections allows observing the relationship between
traffic fluctuation frequency and network performance. In
particular, we have assumed that 15% of the served
connections change their bandwidth during their lifetime.
To increase the fairness of the comparison, the Shift-DAD
policy proposed in section 2.B is considered along with the
First Fit SA scenario for benchmark purposes (i.e., a
scenario where connection re-allocation is allowed). That is,
if it is not possible to accommodate the traffic growth in the
spectrum band originally assigned to a connection, it can be
re-allocated to the first spectrum void with enough capacity.
It is worth to mention that, since avoiding connection re-
allocations is the main purpose of Mid Fit SA, only basic
DAD adaptation policy as in ref [17] is allowed in this case.
In addition, in this paper, we consider BP as one of the
evaluation metrics. It represents the overall failed attempts
to setup new connection requests and bandwidth changes of
already established connections. The 14-Node with 21
bidirectional links NSFnet reference topology has been
selected for the evaluation (Fig. 3). We assume a total
optical spectrum of 1.5 THz per link, which is discretized in
frequency slots of 12.5 GHz. In addition, according to the
asymmetric nature of today's Internet traffic, unidirectional
connections between source and destination nodes are
considered. To appropriately model the offered traffic
granularity (i.e., average number of offered connections per
node and average bit rate per connection), the traffic
generation follows a Poisson distribution process, so that
different offered load values (average number of connection
requests per node) are obtained by keeping the mean
Holding Time (HT) of the connections constant to 200s,
while modifying their Inter-Arrival Time (IAT) accordingly
(i.e., offered load = HT/IAT). Connection bit rates are
randomly generated following a log-normal distribution
over the range from 12.5 Gb/s (1 FS) to 125 Gb/s (10 FSs)
[8]. As mentioned before, for the sake of simplicity, a
spectral efficiency of 1 b/s/Hz has been considered
(realizable with a simple BPSK modulation format).
The first results are shown in Fig.4, where the average
number of offered connections per node ranges from 8 up to
15. As the average bit rate per connection is 35 Gb/s, the
total average traffic generated per node ranges from 280
Gb/s to 525 Gb/s. Concerning the bandwidth variation of the
selected time-varying connections, we assume that their
bandwidth can either be doubled or halved with 50%
probability (for the sake of simplicity). As shown in Fig. 4
(a), the First Fit SA approach with connection re-allocation
outperforms the remainder approaches along the entire
offered traffic range. The notable differences between First
Fit SA with and without connection re-allocation are due to
the fact that with First Fit SA most of the connections need
re-allocation upon bandwidth expansion. Indeed, when the
First Fit SA approach is used, connections are allocated
Fig. 4. a) Network blocking probability for different offered traffic per node values. The average number of offered connections per node
changes from 8 to 15 while the average bit rate per connection is 35 Gb/s, b) Percentage of re-allocated connections vs. offered traffic per
node (only connections requiring expansion are taken into account to compute this percentage).
Khodashenas et al. VOL. 5, NO. 6/JUNE 2013/J. OPT. COMMUN. NETW
6
very close to each other, thus the possibility of finding free
spectral resources around them is very low. As a result,
many of them have to be moved from their original band.
To highlight this, the percentage of re-allocated connections
for each offered traffic value is shown in Fig. 4 (b). As seen,
over 60% of the connections experiencing bandwidth
expansion must be re-allocated when First Fit SA is used.
Therefore, the benefits of First Fit with connection re-
allocation on the BP are achieved at expenses of a large
number of re-allocations, which are both complex and costly
processes. In contrast, an alternative way to achieve
significant benefits with low complexity is the proposed Mid
Fit SA approach. In this case, around one order of
magnitude improvement can even be achieved for low loads
with respect to the First Fit SA without re-allocation, while
no connection re-allocation is needed. Not so pronounced
but still significant benefits are observed for Mid Fit SA
against First Fit SA without re-allocation in highly loaded
network scenarios.
To investigate more about the benefits of our proposal,
similar studies have been done for a fixed offered load of 12
offered connections per node (168 connections offered to the
entire network), while changing the average bit rate
demand per connection from 25 Gb/s to 60 Gb/s (2 to 5 FSs).
According to the mentioned load profile, the average traffic
generated per node ranges from 300 Gb/s to 720 Gb/s. Again
the percentage of time varying connections in the whole
simulation is assumed to be 15%, and the bandwidth of
each randomly selected connection can be either doubled or
halved with 50% probability. As shown in Fig. 5 (a), a
significant reduction in blocking probability can be achieved
by allowing time-varying connections to freely shift in the
spectrum (First Fit SA approach with connection re-
allocation). However, by increasing the average bit rate per
connection, the possibility of finding enough spectrum
resources for re-allocating active connections is reduced. In
fact, for average bit rate values greater than 50 Gb/s the
Mid Fit SA approach leads to better BP performance in the
network. Moreover, this performance is achieved in a
simpler and more cost-effective manner, since no control
plane-driven re-allocation is triggered. Fig. 5 (b) shows the
percentage of re-allocated connections for each average bit
rate value. As shown, by increasing the connection bit rate,
the percentage of re-allocations connections decreases,
which verifies the abovementioned effect.
In the next study, the impact of the percentage of time-
varying connections on the EON performance is evaluated.
The offered load is fixed to 12 connections per node and the
Fig. 5. a) Network blocking probability for different average bit rate per connection values. The average number of offered connections per
node is fixed to 12, while the average bit rate per connection changes from 25 Gb/s to 60 Gb/s, b) Percentage of re-allocated connections vs.
average bit rate per connection (only connections requiring expansion are taken into account to compute this percentage).
Fig. 6. Network blocking probability vs. perce ntage of time-varying
connections.
Fig. 7. Network blocking probability vs. total spectrum per link.
Khodashenas et al. VOL. 5, NO. 6/JUNE 2013/J. OPT. COMMUN. NETW
7
average bit rate demand per connection equal to 35 Gb/s. As
illustrated in Fig. 6, Mid Fit SA improves the performance
of the network in terms of BP in the whole range of the
simulation when compared to First Fit SA without re-
allocation. Considering First Fit SA with re-allocation, as
illustrated, its efficacy in serving time-varying connections
decreases as traffic fluctuation occurs more frequently in
the network. For percentages greater than 30% the Mid Fit
SA approach leads to better BP performance in the network.
Indeed, by increasing the number of time-varying
connections in the network, the possibility to successfully
re-allocate a connection decreases. Therefore, worse BP
performance using First Fit SA with re-allocation against
Mid Fit SA is observed.
Looking for a way to approach the performance of Mid Fit
SA to that of First Fit SA with connection re-allocation, we
target at providing some extra spectrum to the Mid Fit SA
scenario. This would allow finding larger voids and, thus,
decreasing the BP due to lack of spectrum upon bandwidth
expansion. It is noteworthy here that a strategy like this
does not necessarily entail higher network CAPEX, as a
network operator may be underutilizing the entire 5 THz C-
Band bandwidth. Furthermore, if the number of offered
connections to the network remains constant, the number of
devices that must be equipped in the network nodes (e.g.,
transponders) also remains unaltered. For this study, we
consider an average number of offered connections per node
equal to 20, with an average bit rate demand of 35 Gb/s.
Moreover, the percentage of time-varying connections in the
whole simulation is assumed to be 15%, and the bandwidth
of each selected time-varying connection can either be
doubled or halved with 50% probability.
As shown in Fig. 7, such an amount of traffic leads to 1%
blocking probability with First Fit SA with re-allocation if
the total bandwidth per link is 1.5 THz. In contrast, BP
rises up to 3% when Mid Fit SA is applied. Nevertheless, by
increasing the spectrum from 1.5 THz to 1.95 THz the
performance of Mid Fit SA shows no penalty compared to
First Fit SA with re-allocation. Hence, increasing the
operational bandwidth of the fiber links by around 25% to
30% allows Mid Fit SA to achieve the same BP performance
as First Fit SA, but without any connection re-allocation,
which simplifies the network operation to a large extent. In
view of this result, Mid Fit SA becomes an interesting
option for EON operators that can reduce network
complexity by dedicating some extra spectrum in their
potentially underutilized fiber links.
V. CONCLUSION
Elastic optical network (EON) technologies arise as
promising solutions for future high-speed optical
transmission, since they can provide superior flexibility and
scalability in spectrum allocation towards the seamless
support of diverse services along with the rapid growth of
Internet traffic. In this paper, we focused on lightpath
adaptation under variable traffic demands in EONs. The
possibility of utilizing spectral fragmentation for increasing
elastic spectrum allocation capability of EON has been
explored. We proposed a heuristic Spectrum Allocation (SA)
algorithm, called Mid Fit approach, to intentionally
increase spectral fragmentation in the network. In this
approach, the spectrum dedicated to a connection is in the
middle of largest possible free spectral void over the route,
providing greater spectral resource between adjacent
connections. These spectral spaces are used for dynamic
expansions of lightpaths, so as to adapt them to the time-
varying required transmission rate. By means of
simulation, it has been demonstrated that such a proposal
can serve time-varying connections in a simple and cost
efficient manner.
ACKNOWLEDGMENT
This work has been supported by the Government of
Catalonia and the European Social Fund through a FI-
AGAUR research scholarship grant and by the Spanish
National project ELASTIC (TEC2011-27310).
REFERENCES
[1] Cisco White Paper, "Cisco Visual Networking Index: Forecast
and Methodology, 2011-2016".
[2] M. Jinno, H. Takara, B. Kozicki, Y. Tsukishima, Y. Sone, and
S. Matsuoka, "Spectrum-Efficient and Scalable Elastic Optical
Path Network: Architecture, Benefits, and Enabling
Technologies," IEEE Commun. Mag., vol. 47, no. 11, pp. 66-73,
Nov. 2009.
[3] M. Jinno, H. Takara, B. Kozicki, "Dynamic Optical Mesh
Networks: Drivers, Challenges and Solutions for the Future,"
in Proc. ECOC, Sep 2009, 7.7.4.
[4] K. Christodoulopoulos, I. Tomkos, E. Varvarigos, "Dynamic
Bandwidth Allocation in Flexible OFDM-based Optical
Networks," in Proc. OFC, 2011, OTuI5.
[5] T. Takagi, H. Hasegawa, K. Sato, Y. Sone, B. Kozicki, A.
Hirano, M. Jinno, "Dynamic Routing and Frequency Slot
Assignment for Elastic Path Networks that Adopt Distance
Adaptive Modulation," in Proc. OFC, 2011, OTuI7.
[6] A. Castro, L. Velasco, M. Ruiz, M. Klinkowski, J. Fernández-
Palacios, D. Careglio, "Dynamic Routing and Spectrum (Re)
Allocation in Future Flexgrid Optical Networks," Computer
Networks, vol. 56, no. 12, pp. 2869-2883, 2012.
[7] A. Patel, P. Ji, J. Jue, and T. Wang, "Routing, Wavelength
Assignment, and Spectrum Allocation in Transparent Flexible
Optical WDM (FWDM) Networks," in Proc. Photonics in
Switching, 2010, PDPWG1.
[8] O. Gerstel, M. Jinno, A. Lord, B. Yoo, "Elastic Optical
Networking: A New Dawn for the Optical Layer," IEEE
Commun. Mag., vol. 50, no. 2, pp. s12-s20, February 2012.
[9] P. S. Khodashenas, J. Comellas, J. Perelló, S. Spadaro,
"Correlation Between Traffic Granularity and
Defragmentation Periodicity in Elastic Optical Networks," to
be published in Transactions on Emerging
Telecommunications Technologies, 2014.
[10] K. Christodoulopoulos, I. Tomkos, E. A. Varvarigos, "Routing
and Spectrum Allocation in OFDM-based Optical Networks
with Elastic Bandwidth Allocation", in Proc. IEEE Globecom,
2010, ONS06.
[11] T. Takagi, H. Hasegawa, K. Sato, T. Tanaka, B. Kozicki, Y.
Sone, M. Jinno, "Algorithms for Maximizing Spectrum
Efficiency in Elastic Optical Path Networks that Adopt
Distance Adaptive Modulation," in Proc. ECOC, 2010,
We.8.D.5.
[12] K. Christodoulopoulos, I. Tomkos, E. Varvarigos, "Spectrally /
Bitrate Flexible Optical Network Planning," in Proc. ECOC,
2010, We.8.D.3.
[13] K. Christodoulopoulos, I. Tomkos, E. Varvarigos, "Elastic
Bandwidth Allocation in Flexible OFDM-based Optical
Networks," Journal of Lightwave Technology, vol. 29, no. 9,
pp. 1354-1366, 2011.
Khodashenas et al. VOL. 5, NO. 6/JUNE 2013/J. OPT. COMMUN. NETW
8
[14] A. Patel, P. Ji, J. Jue, and T. Wang, "Defragmentation of
Transparent Flexible Optical WDM (FWDM) Networks," in
Proc. OFC, 2011, OTuI8.
[15] K. Christodoulopoulos, E. Varvarigos, "Routing and Spectrum
Allocation Policies for Time-Varying Traffic in Flexible Optical
Networks," in Proc. ONDM, pp. 1-6, 2012.
[16] K. Christodoulopoulos, I. Tomkos, E. Varvarigos, "Time-
Varying Spectrum Allocation Policies and Blocking Analysis in
Flexible Optical Networks," Journal of Selected Areas in
Communications, vol. 31, no. 1, pp. 13-25, 2013.
[17] E. Palkopoulou, I. Stiakogiannakis, D. Klonidis, K.
Christodoulopoulos, E. Varvarigos, O. Gerstel, and I. Tomkos,
"Dynamic Cooperative Spectrum Sharing in Elastic
Networks", in Proc. OFC, March 2013, OTu3A.
[18] F. Cugini, M. Secondini, N. Sambo, G. Bottari, G. Bruno, P.
Iovanna, and P. Castoldi, "Push-Pull Technique for
Defragmentation in Flexible Optical Networks," in Proc. OFC,
2012, JTh2A.40.
[19] R. Wang, B. Mukherjee, "Provisioning in Elastic Optical
Networks with Non-Disruptive Defragmentation," Journal of
Lightwave Technology, vol. 31, no. 15, pp. 223-228, 2013.
[20] A. Rosa, C. Cavdar, S. Carvalho, J. Costa, L. Wosinska,
"Spectrum Allocation Policy Modeling for Elastic Optical
Networks," in Proc. HONET, pp. 242-246, 2012.
Pouria Sayyad Khodashenas received his
B.Sc. degree in 2005 from the University of
Guilan, Rasht, Iran, and his M.Sc. degree in
2008 from University of Tabriz, Tabriz, Iran,
both in electrical engineering. Since October
2009, he has been working towards a Ph.D.
in telecommunications engineering at
Universitat Politècnica de Catalunya (UPC,
BarcelonaTech), Barcelona, Spain, being
advised by Prof. Jaume Comellas. His
research interests include all-optical
communication systems, all-optical switches, nonlinear optics,
quantum optics, multi-layer structures and semiconductor lasers,
especially VCSEL.
Jaume Comellas received the M.S. (1993)
and Ph.D. (1999) degrees in
Telecommunications Engineering from
Universitat Politècnica de Catalunya (UPC),
Spain. His current research interests focus
on IP over WDM networking topics. He has
participated in many research projects
funded by the Spanish government and the
European Commission. He has co-authored
more than 100 research articles in
international journals and conferences. He
is an associate professor at the Signal Theory and Communications
Department of UPC.
Salvatore Spadaro received the M.Sc.
(2000) and the Ph.D. (2005) degrees in
Telecommunications Engineering from
Universitat Politècnica de Catalunya
(UPC). He also received the Dr. Ing. degree
in Electrical Engineering from Politecnico
di Torino (2000). He is currently Associate
Professor in the Optical Communications
group of the Signal Theory and
Communications Dept. of UPC. Since 2000
he is a staff member of the Advanced
Broadband Communications Center (CCABA) of UPC. His research
interests are in the fields of all-optical networks with emphasis on
network control and management, resilience and network
virtualization.
Jordi Perelló received his M.Sc. and
Ph.D. degrees in telecommunications
engineering in 2005 and 2009,
respectively, from UPC. Currently, he is
an assistant professor in
the Computer Architecture Department
(DAC) at UPC. He has participated in
various IST FP-6 and FP-7 European
research projects, such as FP-7 projects
STRONGEST, EULER, DICONET or NoE
BONE, as well as FP-6 projects NOBEL
Phase 2, e-Photon/One+ or COST Action 291. He has published
more than 60 articles in international journals and conferences.
His research interests concern resource management, quality-of-
service issues, and survivability of future optical transport
networks.
Gabriel Junyent graduated in
telecommunication engineering from the
Universidad Politécnica de Madrid,
Spain, in 1973, and received the Ph.D.
degree in communications from the
Universitat Politècnica de Catalunya
(UPC), Barcelona, Spain, in 1979. From
1973 to 1989, he was a Teaching
Assistant and Associate Professor at
UPC, where he has been a Full
Professor since 1989. He founded the Advanced Broadband
Communications Center (CCABA) at UPC in 1990. In the last 15
years he has participated in more than 30 national and
international R&D projects. He is coauthor of more than 150
journal and conference papers.
A preview of this full-text is provided by Optica Publishing Group.
Content available from Journal of Optical Communications and Networking
This content is subject to copyright. Terms and conditions apply.
