1388 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 14, NO. 6, DECEMBER 2006
Generalized Sharing in Survivable Optical Networks
Maher Ali, Senior Member, IEEE
Abstract—Shared path protection has been demonstrated to be
a very efficient survivability scheme for optical networking. In this
scheme, multiple backup paths can share a given optical channel if
their corresponding primary routes are not expected to fail simul-
taneously. The focus in this area has been the optimization of the
total channels (i.e., bandwidth) provisioned in the network through
the intelligent routing of primary and backup routes. In this work,
we extend the current path protection sharing scheme and intro-
duce the Generalized Sharing Concept. In this concept, we allow for
additional sharing of important node devices. These node devices
(e.g., optical–electronic–optical regenerators (OEOs), pure all-op-
tical converters, etc.) constitute the dominant cost factor in an op-
tical backbone network and the reduction of their number is of
paramount importance. For demonstration purposes, we extend
the concept of
1:N shared path protection to allow for the sharing
of electronic regenerators needed for coping with optical transmis-
sion impairments. Both design and control plane issues are dis-
cussed through numerical examples. Considerable cost reductions
in electronic budget are demonstrated.
Index Terms—Optical networks, shared protection.
I. GENERALIZED SHARING OF
NETWORK RESOURCES
S
HARED path protection has been established as a mech-
anism for providing considerable savings in terms of the
number of provisioned wavelengths [1]–[6]. In this protection
scheme, optical services whose primary routes are physically di-
verse, can be allowed to share protection wavelengths (on their
protection paths) on some of the fiber links. This can be ac-
complished since, under the single-link failure scenario, physi-
cally diverse primary routes will not fail at the same time; hence
their protection paths will not be activated simultaneously. Pre-
vious work on shared protection considered the wavelength (or
bandwidth) as the only shared resource that needs to be opti-
mized. This focus ignores the fact that the dominant cost of op-
tical backbone networks is that of the optical–electronic–optical
(OEO) devices [1] that are needed for dealing with optical trans-
mission impairments and/or realization of wavelength conver-
sion [10]. In this work, we introduce the
Generalized Sharing
Concept which allows for the sharing of not only fiber resources,
but also valuable node equipment. In this context, lightpaths
[11] can share wavelengths and regeneration points on their pro-
tection paths. Motivated by this new sharing concept, we study
the optimization of provisioned OEOs in a survivable optical
network. In such a network, a lightpath is routed all-optically
as far as possible and it encounters regeneration at intermediate
nodes when the transmission impairments significantly degrade
Manuscript received December 2, 2003; revised June 8, 2005; approved by
IEEE/ACM T
RANSACTIONS ON NETWORKING Editor C. Qiao.
The author is with Alcatel Research and Innovation, Plano, TX 75075 USA
(e-mail: maher.ali@alcatel.com).
Digital Object Identifier 10.1109/TNET.2006.886293
Fig. 1. Example illustrating the generalized sharing concept. IP routers are in-
terconnected via switching nodes. Each switching node is capable of all-optical
switching using a photonic cross-connect (PXC) and optional OEOs used in re-
generation/wavelength conversion. The example shows two backup lightpaths
sharing an OEO at Node
X
. In addition,
on Fiber
L
is shared by two backup
lightpaths.
the signal [12], [13]. After that, we address some of the impli-
cations on the control plane.
A. Illustration of the Concept
Fig. 1 illustrates the idea behind the generalized sharing con-
cept. Three protected lightpaths are established:
, ,
and
. Since the primary routes do not have physical links
in common, the resource(s) on their respective backup paths can
be shared. In the figure, we notice that the backup paths for con-
nections
and share an OEO at Node . When the
primary of either lightpaths fails, their corresponding backup
path is activated and it can use the regeneration node. Without
this sharing, the two backup paths, which each require an OEO
at Node
to clean up their signal from accumulated optical
impairments, may potentially need to have a dedicated OEO.
The other shared resource is the wavelength and is shown in the
figure where Channel
on Link is shared between connec-
tions
and . In general, backup routes can share
the same resource if their primary routes are physically diverse.
For every shared object in the network, a sharing table must
be employed. This sharing table contains an identification of
the object being shared as well as a list of fibers’ identification
numbers. This list is formed by the union of the fibers traversed
by the primary paths of all services whose backup routes share
that object. Every backup path that wants to share an object must
have the property that its primary route does not traverse any of
the fibers of that object’s sharing table. We utilize two sharing
1063-6692/$20.00 © 2006 IEEE
ALI: GENERALIZED SHARING IN SURVIVABLE OPTICAL NETWORKS 1389
Fig. 2. Illustration of Level 1 sharing. The two backup paths share a fiber termi-
nated by an OEO regenerator at Node 4. The network example shows IP routers
interconnected via a mesh of switching nodes. Each switching node is com-
prised of a photonic cross-connect (PXC), OEOs, and multiplexers.
tables types: a sharing table of the channels and a sharing table
of the OEOs.
It is worth noting that no additional device is needed for the
realization of OEO sharing, since as it will be shown, the optical
switching fabric directs the specific backup signal to a pool of
shared OEOs when failure occurs.
In the following, we provide details on how the sharing
scheme extended to OEOs can be realized.
B. Two Levels For Sharing a Regenerator
In practice, one can think of two sharing levels (schemes) that
govern the sharing of OEO devices at intermediate switching
nodes. The first scheme (Level 1 sharing) is an extension of the
currently used wavelength-sharing scheme used in bandwidth
optimization. In this scheme, a channel, say
on Fiber ,
requires the reservation of an OEO device at Node
if one
or more of the protection paths sharing Channel
requires
OEO regeneration. This requirement for OEO regeneration is
dictated by the need for signal recovery, due to accumulated
optical impairments, and/or the need for wavelength conver-
sion, due to the unavailability of Channel
on the next hop.
In either case, once it is decided that one of the protection
services needs an OEO at a node, all new protection services
that share that channel can use this OEO device with zero
additional cost. Fig. 2 shows an illustration of Level 1 sharing.
In the example, two fiber-disjoint primary paths are estab-
lished: 1)
,
and 2)
.
Under single link-failure scenario, one observes that since the
lightpaths are fiber-disjoint, both primary lightpaths cannot fail
at the same time; hence at most one of their protection light-
paths will be activated. In the figure, both protection paths are
routed on
. Let us assume that the
transmission quality of the transmission system between Nodes
Fig. 3. Illustration of Level 2 sharing. The two backup paths share a regenerator
at intermediate Node 4.
1 and 4 is poor that it requires signal regeneration at Node 4. In
this case, one OEO device can be reserved to accommodate up
to
(in this case ) protection paths.
The second scheme for sharing OEO devices is Level 2
sharing. In Level 2, two protection paths whose primary
paths are fiber-disjoint can share an OEO device at a
given intermediate node irrespective of the bandwidth
(channel) sharing. Fig. 3 illustrates the idea. Two primary
lightpaths
and are
fiber-disjoint. Their respective protection paths are
and
share an
OEO device at Node 4 even though the two input signals at
Node 4 come from two distinct input fibers and exit at two
different fibers. At the time of the activation of either protection
signal, the signal is diverted to an OEO bank where it receives
regeneration and then it is switched back in the optical domain
to its output fiber. Thus, in this architecture, routes which
do not share any fiber on their backups can still share OEOs
on intermediate nodes. This allows for maximum sharing
optimization.
Clearly, if the input nodal degree of every node in the network
is equal to 1, Level 2 degenerates to Level 1. However, real life
networks have input nodal degree that is greater or equal to 2
allowing for the utilization of this intra-fiber OEO sharing as it
will be demonstrated in numerical examples.
1390 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 14, NO. 6, DECEMBER 2006
C. Basic Optimization Problem
In the following, we give a definition of the basic optimization
problem discussed in this work. This problem will be used for
both the design and protocol issues discussed in the following
sections.
Definition Pair of Disjoint Paths with Minimum Total OEOs
(PDP-OEO) Problem
The PDP-OEO problem is defined as follows. Given: a) a
set of nodes and a set of fiber links connecting these nodes,
b) wavelength availability on fiber links and sharing tables on
every node, and c) a unidirectional connection request from
to . The objective is to find two fiber-disjoint paths between
and such that the total number of regenerators needed to cope
with transmission impairments is minimized.
It is worth noting that in this work we focus primarily on
transmission impairments as the only need for OEOs. However,
the methodology developed in this work can be equally applied
to wavelength converters needed for coping with the wavelength
continuity constraint. This later problem is left for future re-
search. In the following section, we provide a solution for the
above problem and use that solution for the embedding of a vir-
tual topology (i.e., set of demands) on a given physical topology
with minimum number of OEOs. This solution is also used for
the dynamic provisioning of lightpath services.
II. N
ETWORK DESIGN
PERFORMANCE STUDY
In this section, we focus on the design problem motivated by
the generalized sharing concept. Informally, given the network
physical topology and a set of network connections representing
the virtual topology required, the objective is establish all con-
nections with their protection paths such that the total number of
OEOs in the network is minimized. In the following, we discuss
some design algorithms that can be used to optimize the network
cost. Next, we provide a simple example illustrating this issue.
After that, a more realistic network is considered.
A. Network Design Issues
The network design problem is given by the following:
Definition: Survivable Network Design With Minimum OEOs
(SND-OEO).
Given: 1) a network physical topology represented by a graph
composed of the set of nodes and the set of fibers
connecting these nodes with the set of available wavelengths
on a fiber given by
, and 2) a set of demand requests , where
is a unidirectional demand request for a light-
path from
to . The objective is to minimize the total number
of OEOs used in routing both primary and backup paths for the
demand in
. The pseudo-code in Fig. 4 provides a heuristic for
this problem.
The heuristic optimizes the OEO resources in two phases.
In phase I (H-Basic), a sequential establishment of sessions is
performed. As sessions are established, more opportunities for
sharing OEOs become available. The main component of this
phase is the
function shown in Fig. 5.
employs Best-Fit (BF) heuristic for
the establishment of the primary and backup lightpaths. It goes
Fig. 4. Network design algorithm.
through all different wavelengths for the primary
and backup paths and finds the two routes for primary and
backup using Wavelengths
and , respectively. In testing
each
and pair, we attempt first at finding the primary path
and then finding the backup path. The pseudo-code in Fig. 6
finds the cost for establishing the primary path.
For finding the primary path, the algorithm constructs a graph
such that two nodes are connected iff Wavelength is avail-
able. It labels each edge of the graph such that fibers with higher
impairments and/or more shared channels have less chance of
being picked up by the routing algorithm. We achieve this by
labeling an Edge
with ,
where
is the amount of impairments on Fiber and
is the number of wavelengths used for
backup on
. By this formula, we hope that the shortest path
algorithm avoids using shared resources (e.g., wavelengths)
that may be helpful in minimizing the cost for the backup path.
Dijkstra’s shortest path algorithm is run on this new graph to
find the shortest path from
to .
Let
be the sharing table for the wave-
length
on link . In other words, it holds the set of all links
in all primary routes whose backup routes use (i.e., share)
wavelength
on link . Let be the
sharing table for the OEO of wavelength
at node . Similar
to the case of the wavelength sharing table, it contains all links
in all primary routes whose backup routes use (i.e., share) the
OEO of wavelength
at node . After finding the primary
path, we aim at finding the backup path (Fig. 7). This backup
path can only use edges that are not used by the primary path.
In addition, if Wavelength
is being used for backup and
the primary path shares one or more fibers with the sharing
table of Wavelength
on , the edge is excluded from the
computation. Each edge
is labeled with 0 if an OEO
is available for free use at Node
. Otherwise, is labeled by
. The labeling is used to favor fibers ending with a free
regenerator and at the same time minimize the impairments.
Dijkstra’s shortest path algorithm is run on this new graph to
find the shortest path from
to . If such a path exists, we
determine the placement of regeneration points at intermediate
nodes as follows. Using
and starting
with the ingress node and moving downstream on the path to-
wards egress node, the amount of accumulated impairments is
recorded as well as the location of the last regenerator that can
be shared (here differences between Levels 1 and 2 determine
whether an OEO can be shared or not).
ALI: GENERALIZED SHARING IN SURVIVABLE OPTICAL NETWORKS 1391
Fig. 5. Establish a session heuristic.
Fig. 6. Cost of establishing primary path for a session.
Whenever the impairments reach the threshold, the last
shared regenerator is used to clean up the signal. If no such
regenerator can be shared, a new regenerator is installed. The
process is repeated until we reach the egress node. Total cost is
composed of only newly added regenerators.
It is worth noting that this sequential approach (i.e., fixing
a primary route and then looking for the backup) may some-
time result in a trap situation, in which a pair of diverse pair of
primary and backup paths cannot be found, even if one exists,
because the primary was not correctly selected. Finding an op-
timum solution is unfortunately an intractable problem, due to
the fact that different cost-metrics are used for computing the
primary and the backup. For more information on this problem
and possible remedies, please refer to [14] and [15].
The time complexity for establishing one session is
assuming the shortest path is computed in
and labeling an auxiliary graph and installing regenerators is
done in
. For larger values of , one can replace the
Best-Fit approach with a scheme that is not as computation-
ally-intensive (such as First-Fit) to achieve time complexity of
.
After the sequential establishment of the sessions, we use the
second phase to further optimize the resources. In Phase II, the
well-known Hill-Climbing (HC) algorithm is used to optimize
the solution in Phase I. At each step of the HC heuristic, all ses-
sions are inspected at a time. For each session, we find its cur-
rent cost composed of OEOs on primary and backup routes. The
session is then removed and resources are freed. Next, the ses-
sion is established by taking advantage of all sharing resources
available. If the new cost (which is computed as in Phase I) is
less than the previous cost, a profit is made. If the current profit
is larger than the profit achieved so far, the session becomes the
candidate for re-routing. At the end of inspecting all sessions,
the candidate session whose re-routing provides the maximum
profit is re-routed. This step is repeated until there is no candi-
date session for re-routing (i.e., no further enhancements to the
number of OEOs provisioned). Fig. 8 shows a pseudo-code for
the HC heuristic.
B. A Simple Example
In the following, we present an example showing both the
benefits of shared protection compared with 1+1 dedicated path
protection and the cost reduction using Level 2 sharing. Con-
sider a simple physical topology of an 8-node network as shown
in Fig. 9. The network consists of optical cross-connects (pho-
tonic switches with regeneration capability) connected via 13
1392 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 14, NO. 6, DECEMBER 2006
Fig. 7. Cost of establishing backup path for a session.
Fig. 8. Hill-climbing heuristic.
bi-directional fibers. Every switching node is assumed to have
a client (e.g., an IP router) connected to it. Consider every fiber
carries up to 16 channels. Also, consider that every fiber uti-
lizes Long Haul (LH) transmission system with maximum op-
tical reach of 400 km. All links are assumed to be of 300 km
length; thus requiring regeneration at every intermediate hop.
Consider a mesh virtual topology is requested to be imple-
mented on top of this physical network. In other words, traffic
demand is composed of one lightpath request from every node
to every other node with 56 connection requests. We utilize
three schemes for optimization of the virtual topology cost dom-
inated by OEO devices required to clean up the signal. The first
scheme is a sequential application of an optimal solution for
the problem of finding a pair of fiber-disjoint paths with min-
imum number of total regeneration on these routes given the
current configuration of the network. The optimal solution is
found by using an integer-linear program (ILP) shown in the
appendix. This ILP program is sequentially solved for all light-
path requests where the solution of the current request utilizes
the shared resources of all previous request and is also con-
strained by their routing solutions. Two ILP solutions are used:
ILP-Level 1 and ILP-Level 2 for the two sharing levels. The
second scheme uses the H-Basic heuristic discussed in the pre-
vious section. Note that the H-Basic heuristic is similar to the
