Carrier Ethernet Over SDH
Transcript of Carrier Ethernet Over SDH
-
8/2/2019 Carrier Ethernet Over SDH
1/6
Carrier Grade Ethernet versus SDH
in Optical Networks: Planning Methods
and CAPEX Comparisons
Domenico Siracusa, Guido MaierDepartment of Electronics and Information, Politecnico di Milano, Via Ponzio 34-35, 20121 Milan, Italy
Email: {siracusa,maier}@elet.polimi.it
Abstract The new Carrier Ethernet services offered byproviders require not only a huge amount of bandwidth, but alsoa more flexible access to bandwidth that the traditional trans-port networks, based on circuit-switching, can hardly provide.Carriers have nowadays the option of migrating to new layer-2frame-switched technologies developed ad hoc to support CarrierEthernet. An important question is whether this migration isreally cost effective. This paper deals with the problem of de-
signing a transport network able to support Carrier Ethernet. Wecompare two network models, one based on layer-2 switching andEthernet interfaces, the other based on legacy circuit switchingand SDH interfaces. Minimization of investments for networkinterfaces is carried out for the two models, given the same static-traffic matrix to be supported. Both an ILP formulation and aheuristic method are proposed to solve the design problem. Theresults presented for comparison are obtained by applying theoptimization procedures to a case-study network.
Index Terms Optical networks, network planning, CarrierGrade Ethernet
I. INTRODUCTION
In recent years providers are demanding for a new gen-eration of IP networks able to face the challenges posed
by new services and increasing bandwidth demand. This
main driving force is fostering the current evolution of the
transport-network carriers. The traditional transport infrastruc-
ture, almost-entirely based on large-capacity circuit-switched
trunks, does not seem appropriate any more to deliver packet-
based traffic. SONET/SDH interfaces, ubiquitously deployed
until few years ago, are today matched by the Ethernet-
family interfaces. Thus, the most interesting option available to
carrier operators is the migration from a SONET/SDH-based
infrastructure to the apparently less-expensive and packet-
switching oriented Ethernet counterpart.
The Metro Ethernet Forum (MEF) has defined the services
and requirements of the Carrier Ethernet (CE) [1]. In order to
provide services with their QoS attributes, an Ethernet Virtual
Connection (EVC) [1], a sort of tunnel between two or more
endpoints of a network, has to be established. The delivery
of the end-to-end Ethernet service must be transparent to the
(potentially multiple) transport technologies adopted in the
The work presented in this paper was carried out with the support of theBONE-project (Building the Future Optical Network in Europe), a Networkof Excellence funded by the European Commission through the 7th ICT-Framework Programme.
networks crossed by the EVC. This leaves open the choice
of the technology to the carrier operators.
The currently most common transport solutions to sup-
port CE service are: Ethernet-over-SONET/SDH (EoS) and
IP/MPLS tunneling. Both solutions make use of well-
developed and established switching equipment respectively
operating at layer-1 (circuit-switching digital cross connects)
and at layer-3 (label/packet switching routers).
The activity of standardization of the MEF has spurred ma-
jor standardization organizations, such as IEEE, ITU and IETF,
to develop standards for brand-new network technologies
supporting CE through layer-2 switching. In particular, IEEE
recently standardized Provider Backbone Bridging - Traffic
Engineering (PBB-TE) [2], while ITU-T developed Transport-
MPLS (T-MPLS) [3]. After concerning about the full compati-
bility of T-MPLS with already established IP/MPLS standards,
a liaison between ITU-T and IETF is leading to the definition
of MPLS-Transport Profile (MPLS-TP).
In the next years, we will see which one of the two options,
whether to continue with legacy equipment or to migrate to thenew layer-2 switching concept, will prevail upon the carriers.
Surely, the final choice will be made on the basis of the cost
and the performance of the various solutions.
The main objective of this work is to make a comparison
between the CE transport solution based on Packet-over-SDH
and the upcoming layer-2-switching CE technologies. This
preliminary paper is dedicated to off-line network planning and
optimization on the basis of a static-traffic matrix. In particular,
we will focus on the problem of minimizing the number of
deployed node-interfaces (SDH and Ethernet in the two cases,
respectively). We provide a mathematical formulation of the
network design problem and compare the results given bydifferent optimization algorithms (ILP and heuristics) in a
case-study network. Our intent is to propose new CE network-
design methods and principles. Beside that, we would like to
numerically measure (by simulation) how much the bandwidth
efficiency and multipath capability combine to affect the
choice between Ethernet or SDH technologies.
From the standpoint of this paper, for the assumptions later
reported, the PBB-TE and T-MPLS technologies are treated
as similar. Therefore, we will refer to a more general layer-2
packet-switched (or better frame-switched) CE implementation
instead of explicitly referring to any of such technologies.
978-1-4244-6404-3/10/$26.00 2010 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings
-
8/2/2019 Carrier Ethernet Over SDH
2/6
This article is structured as follows: after a short description
of prior works (Section II), we discuss in Section III the
issue of planning a CE network in the two SDH and layer-2-
switched implementations, pointing out some assumptions. In
Section IV, we introduce the design and optimization problem
providing a mathematical ILP formulation. In Section V, we
present a simulated-annealing heuristic method. In Section VI,
we show the comparison between the different transport tech-nologies. Section VII draws some conclusions.
II . PREVIOUS WORKS
The main objective of this work is to make a compari-
son between transport technologies used to support CE. The
number of papers currently in literature comparing either the
different CE implementations or legacy transport to them is
rather limited.
Different state of the art papers have been presented, from
[4] presented in 2005 which examines CE technologies, to the
papers based on the CE solutions, such as [5] and [6] or [7].
In June 2006, a technical and economical analysis of PBB
transport technology was presented [8]. A similar analysis
was presented by [9]. A paper concerning the design of a CE
network appeared in IEEE Journal of Lightwave Technology
in January 2008 [10]. The authors studied the problem of
designing reliable and cost-efficient high-rate carrier-grade
Ethernet in a multiline-rate optical network under signal
transmission-range constraints. The same authors presented
other two articles: [11] and [12]. The first one offers an optimal
solution of the transmission-range problem by adopting ILP.
In the second work, the authors propose an algorithm that
provides better performance than the shortest-path one.
In May 2008 [13] proposes a control plane for Carrier Grade
Ethernet networks, taking into account the T-MPLS transporttechnology. The paper [14] describes how to use SDH infras-
tructure to deliver CE services from the providers perspective.
In September 2008, IEEE Communication Magazine dedicated
a special issue to CE technology: in [15], the work done by [7]
is extended; in [16], the attention is focused on the comparison
between PBB-TE and SDH technologies from a qualitative
point of view.
During 2009 several articles regarding Carrier Ethernet
technologies have been published. Among them [17] which
makes a comparison between T-MPLS/MPLS-TP and PBB-TE
from the application point of view and [18] which investigates
the topology of IP layers that minimizes the total transportnetwork cost upon a realistic reference network.
III. NETWORK MODELS AND DESIGN-PROBLEM
DEFINITION
Let us better define the network scenarios we are consider-
ing in this comparative work. From now on, we will refer to
layer-2 packet-switched transport technologies as Packet and
to SDH transport technologies as SDH.
There are several reasons for a possible migration of the
carrier network infrastructure from the SDH to the Packet
technology (Section II). SDH interfaces are regarded as being
more expensive. Moreover, with Packet technology the rate
of flows assigned to each EVC varies with continuity, while
SDH sets-up circuits with fixed bandwidth granularity. We
do not consider here other potential advantages of the Packet
solutions, such as statistical multiplexing, or other QoS related
issues such as delay, delay variation, etc. Such aspects will
be analyzed in future extensions of the paper. Moreover, this
paper considers only point-to-point EVCs (E-Line service),leaving multicast connections (E-Lan and E-Tree service) to
future studies.
Since standardization of the control plane is not yet com-
plete for all the CE technologies, in the Packet case we have
to make some simplifying assumptions on the behavior of
the control of routing operations. In particular, we assume
that all the packets (frames) of an EVC are forwarded as
a single flow between two endpoints on an unique route
through the network, with no bifurcations or merging points.
This assumption complies with the examples provided also in
standards and in [15]. In the SDH case, thanks to the Virtual
Concatenation (VCAT) technology, the traffic of an EVC can
be transported on different Virtual Container (VC) paths which
can follow different routes through the network.
In order to provide a meaningful comparison, we consider
only 10 Gbit/s Ethernet (transmission bit rate of 10312.5
Mbit/s and capacity for PDU payload of 10000 Mbit/s) and
STM-64 (transmission bit rate of 9953.28 Mbit/s and capacity
for PDU payload 9584.64 Mbit/s) interfaces, respectively for
Packet and SDH.
In the Packet case, any value of bandwidth can be reserved
for each EVC on each link of the network. To avoid limitations
in the transmission distances (link lengths), we suppose that
the Ethernet signals generated by the interfaces are trans-
parently carried by Optical Transport Network (OTN) [19].These interfaces allow to extend the reach of transmission
of Ethernet systems without requiring any intermediate en-
capsulation protocol or any change to frame formats or line
encoding (64B/66B). In particular, the so called 10GBASE-R
solution includes the Ethernet frames directly into a slightly
over-clocked version of standard Optical Data Unit (ODU)
container: the ODU-2e container.
In the SDH case, the EVCs have to be accommodated into
the SDH paths. We assume that a quantum of bandwidth
for SDH corresponds to a VC-4 container (payload 149.760
Mbit/s), which thus represents the minimum granularity for
bandwidth reservation in the circuit-switched case. Accordingto a popular and effective EoS implementation, we further
assume that Ethernet frames are encapsulated into the VC-4
containers by the Generic Framing Procedure (GFP) (GFP-F
mapping). As GFP is a very efficient protocol, we will neglect
the limited overhead introduced by frame mapping.
Given all the above assumptions, we can now define the
problem we want to deal with. The network topology (nodes
and links) is given, as well as a set of EVC requests. The
links of the network are optical and contain a variable number
of channels, obtained by multiplying the number of fibers
by the number of wavelengths on each fiber. As nodes are
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings
-
8/2/2019 Carrier Ethernet Over SDH
3/6
assumed to be all opaque (no transparent passthrough allowed,
wavelength conversion and regeneration always available at
each node), the number of wavelengths and fibers exactly used
is irrelevant. The only relevant condition is that the number
of optical channels active on a link in each direction has to
be equal to the number of deployed interfaces. Notice that
actual transmission systems are always symmetrical in the two
propagation directions across a link. Hence, we will have toimpose that the number of interfaces deployed at the two edges
of the same link must be the same.
Each EVC request specifies a source and a destination
among the nodes of the network and a bandwidth to be
reserved for the EVC. EVCs are all unicast and unidirectional
and the traffic matrix is assumed static (permanent connec-
tions). The design occurs offline: nodes have to be equipped
with a suitable number of interfaces, so that enough capacity
exists to accommodate all the EVCs. The function to be
minimized is the total number of interfaces to be deployed. The
output consists of the number and location of the interfaces as
well as the routing of all the EVCs.
In both Packet and SDH cases, we assume that traffic
injected into each EVC is controlled at the source by per-
forming a continuous shaping or policing. The maximum
information rate is strictly enforced, possibly allowing for
short temporary violations to accommodate small-size bursts
of data. The enforced maximum information rate for each EVC
corresponds (after adding frame headers) to the bandwidth
reserved in the network for that EVC. This assumption on
traffic is particularly important for the following reason. One
may wonder if our procedure that minimizes the number of
interfaces is not detrimental for QoS, as it tends to overload the
remaining interfaces. This side effect is avoided if we assume
that statistical multiplexing is not exploited in designing thenetwork in the Packet case, consistently with what previosly
stated. Indeed, this condition is ensured if:
1) each flow loading an EVC is forced to guarantee a
maximum burst-size;
2) an independent portion of buffer (or an independent
buffer) is reserved in each transit node of the network
to each EVC, so that the size of such a buffer is
significatively larger than the corresponding maximum
burst-size.
Under the above hypothesis, a fluid-flow model [20], [21] can
be adopted to analyze the network. According to this model,
the packet loss rate on each EVC (or equivalently the delay)becomes independent of the total load of the links: each EVC
can be loaded up to the capacity reserved in the network
for it, without causing instability. Obviously the advantages
of statistical multiplexing are lost, but this is not a problem.
At the same time, the model allows us to better focus our
Packet vs. SDH comparison just on bandwidth granularity and
interface-related CAPEX.
It should be finally noted that the design procedure pre-
sented in this paper supports EVCs differentiation into two
classes of service: protected and unprotected EVCs. Protected
EVCs requires the reservation of the EVC capacity along a
primary path as well as a secondary one, with the two paths
link-disjoint (only dedicated path protection is considered).
IV. INTEGER LINEAR PROGRAMMING FORMULATION
The formulation we have adopted allows us to apply it to the
two Packet and SDH cases with just a few modifications. Here
follows the description of data, variables, objective function
and constraints: Data:
- G = (V,E): network topology, in which V and E are thesets of nodes and (unidirectional) links, respectively(V=|V|; E=|E|).
- bhs,d: connection request number h of b Mbit/s betweensource node s and destination node d.
- Hs,d: number of requests between s and d, Hs,d = |Hs,d|.- S: set of source nodes.- D: set of destination nodes.- K: set of deployable interfaces per each link.- B: bandwidth of an interface (bandwidth available for
payload).- M: maximum number of interfaces for each link.
Variables:
- Xs,d,hi,j,k
: Xs,d,hi,j,k
= 1 (i = j) if connection request number hfrom s to d is routed across link (i, j) over interfacek, otherwise Xs,d,hi,j,k = 0.
- Ii,j,k: Ii,j,k = 1 (i = j) if the interface k on the link (i, j)has been deployed, otherwise Ii,j,k = 0
Objective Function: route all connection requests deploying theminimum number of bidirectional interfaces.
Minimize :
(i,j)E,kK
Ii,j,k (1)
Constraints:
- Solenoidality constraints, for all s S, d D, and h Hs,d:
jV,kK
Xs,d,hj,i,k =
jV,lK
Xs,d,hi,j,l i = (s, d) (2)
jV,kK
Xs,d,hi,j,k = 1 i S (3)
iV,kK
Xs,d,hi,j,k = 1 j D (4)
- Capacity constraints:
hHs,d
bhi,jXs,d,hi,j,k Ii,j,kB (i, j) V, k K (5)
- Bidirectionality of interfaces constraint:
Ii,j,k = Ij,i,k (i, j) E, k K (6)
Regarding the distinction between Packet and SDH network,we introduce the following rules on the data set:
SDH network:
- B=64 C where C is VC4 capacity.- Each connection request has the same bandwidth as the
payload of a VC4, that is C.- The total number of connection requests is equal to:
Hs,d = bhs,d/C
Packet network:
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings
-
8/2/2019 Carrier Ethernet Over SDH
4/6
- B=10 Gbit/s.
Concerning the protected class of service, we use a dedicatedpath protection algorithm and we introduce a new variable
Ys,d,hi,j,k that is identical to Xs,d,hi,j,k but it refers to the protection
path. Then the capacity constraint becomes:
hHs,d
bhi,j(Xs,d,hi,j,k + Y
s,d,hi,j,k ) Ii,j,kB (i, j) V, k K (7)
The description of Packet and SDH data sets does not differfrom the previous one. Some new constraints must be added tothe unprotected class of service formulation. These constraintshave to be complied for all h Hs,d:
jV,kK
Ys,d,hj,i,k =
jV,kK
Ys,d,hi,j,k i = (s, d) (8)
jV,kK
Ys,d,hi,j,k = 1 i S (9)
iV,kK
Ys,d,hi,j,k = 1 j D (10)
In the end we consider the link-disjointness constraint:
kK
(Xs,d,hi,j,k + Ys,d,hi,j,k ) 1 (i, j) E, h H
s,d(11)
The computational complexity of the ILP in terms of the
number of variables for the unprotected case is equal to
EK[V(V1)H+1]. In the worst case (fully meshed network)it goes as O[V4KH]. For the protected case the number ofvariables is EK[2V(V 1)H+ 1] and, in the worst case, thecomplexity goes as O[2V4KH].
The proof of NP-completeness for the aforementioned prob-
lem () can be given by restriction [22], i.e. by proving that
contains a known NP-complete problem
as a special case.In particular we can restrict the instances of our problem to
a network composed by only two nodes and a link. Under
these conditions our problem corresponds one-to-one to the
bin packing problem, which is NP-hard in the strong sense
[22]. The equivalence (not formally proven here for brevity)
is due to the following properties: the objective function (1),
in which only interfaces are considered as indexes, is the same
of the bin packing problem; constraints (2), (3) and (4) can be
disregarded; capacity constraint (5) has a clear correspondance
to the bin packing capacity constraint; equation (6) only sets
a scaling factor which does not affect the decision problem.
V. HEURISTIC OPTIMIZATION BY SIMULATED ANNEALING
As NP-hard problems typically require long computational
time, we need a heuristic approach to solve realistic-size de-
sign problems. We chose to develop a meta-heuristic approach
based on Simulated Annealing (SA) [23].
We applied it to our case as follows. After a first routing
phase in which all requests are satisfied, an optimization phase
follows, with the purpose of minimizing resource utilization.
The number of deployed Ethernet and SDH interfaces in the
Packet and SDH case is the cost function we are trying to
minimize. The optimization phase uses the Algorithm 1.
Algorithm 1 Simulated Annealing
1) Set the initial temperature (rerouting probability);
2) Save the current state (routing of all connections, de-
ployed interfaces as optimal and current number of
deployed interfaces as minimum cost);
3) Randomly choose a link of the network;
4) Try to reroute all the connection requests passing
through the link: all requests are successfully rerouted: the released
resources (interfaces) can be canceled and a confir-
mation of the positive result is returned;
one or more requests cant be rerouted: the previous
state is restored and a confirmation of the negative
result is returned;
5) Discriminate on the basis of the outcome of the rerouting
routine:
positive result:
- if the number of deployed interfaces decreases
compared to the minimum cost, store the new
state as optimal and keep it for the next SAroutine; go to 6;
- if the number of deployed interfaces increases
compared to the minimum cost but it is under
a threshold, keep the new state for the next SA
routine; go to 6;
- if the number of deployed interfaces increases
compared to the minimum cost and it is over a
threshold, restore the optimal state; go to 6;
negative result: generate a random number between
0 and 1, if this number is lower than the rerouting
probability, add a channel (two interfaces) to all
links, with the exception of the link in which thererouting routine has been attempted;
6) After a fixed number of steps, decrease the temperature
(rerouting probability);
7) Repeat the steps from 3 to 6 until the temperature
(rerouting probability) is below a fixed threshold.
V I. LAYER-2 SWITCHING VERSUS SD H
We now discuss the comparison between layer-2 switching
and SDH in terms of number of deployed interfaces (CAPEX),
which is intended to be the final goal of the paper. On this
account it should be noted that more than providing a response
on which is the most cost effective solution, we intend
to elaborate a methodology of comparison to quantify the
difference between the two technologies in terms of number
of deployed interfaces, trying to identify network and traffic
conditions that make one solution more suitable than the other.
Given the assumptions described so far on the defined
network models and scenarios, two main phenomena affect the
number of deployed interfaces: (a) granularity; (b) multipath.
SDH suffers a granularity impairment compared to layer-2
switching because: i) the bandwidth in SDH can be assigned
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings
-
8/2/2019 Carrier Ethernet Over SDH
5/6
to connections in chunks of VC-4 capacity; ii) the bit rate ofan SDH interface (9953.28 Mbit/s) is slightly lower than the
rate of a 10GE interface (10312.5 Mbit/s). It is easy to realize
that the global effect of the granularity impairment decreases
as the overall volume of traffic and as the average bandwidth
required per traffic flow get large. Viceversa, granularity effect
is strengthened when the average path-length (in number of
hops) increases.On the other hand, the layer-2 switching solution suffers a
multipath impairment compared to SDH. This is due to the
capability of SDH, given by VCAT, of slicing a single flow
on multiple paths across the network, creating a load-sharing
effect that is expected to improve performance. However, it
should be considered that the multipath effect can be actually
appreciated in case two or more paths with disjoint interfaces
exist and the amount of available bandwidth on the minimum
cost path is less than the requested one; this means that con-
nection requests are forced to pass through different disjoint
paths. Thus, low nodal degree and small number of available
resources are expected to reduce the impact of the multipath
impairment, especially when the requested bandwidth per
connection is low.
The neat result of the combination of granularity and
multipath is not obvious. Let us consider the simple case in
which the requested bandwidth is equal for each connection
and let us focus a single network node n. The number ofoutgoing interfaces needed by n in the Packet and SDH casebe OIP and OIS , respectively. We have
OIP = r/BP/b (12)
OIS = (b/CV C4 r CVC4)/BS (13)
where r is the number of connection requests of bandwidth barriving at the node n; CVC4, BP, BS , are the bandwidths ofa VC4 container, a Packet and an SDH interface, respectively.
It should be noted that r accounts for connection requests forwhich the considered node is either the source node or a transit
node.
In figure 1 we show the two curves representing OIP andOIS of n in the simple theoretical case of a network of twonodes n and m and one link ln,m. In such case n is sourceof all the r connection requests (n can not be transit node)and thus r can be assigned arbitrarily (r = 15 in the figure).The two curves are represented as functions of the normalized
bandwidth = b/BP. The vertical axes is also normalizedto the maximum number of output interfaces, which is givenby SDH when is equal to 1 (since BP > BS). It canbe observed that for low values of the two technologieshave similar trends; sometimes SDH needs more interfaces,
because of the granularity impairment. Around intermediate
values of this trend changes, and for 0.5 the SDHtechnology is noticeably cheaper than Packet, because of the
the multipath impairment prevailing. For rational values of
, Packet always performs better than SDH because it fillsexactly the available payload on the interface and the multipath
effect becomes useless, leaving the granularity impairment to
0 0.1 0.2 0.3 0.4 0.50
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Normalized Requested Bandwidth ( )
Normalizednumberofou
tputinterfacespernode
PACKET
SDH
Fig. 1. Number of output interfaces per node needed to setup connectionswith equal requested bandwidth in a point-to-point link.
prevail. We have verified that the same behavior holds for non
degenerative networks, this suggests that estimating a CAPEX
comparison between SDH and Packet by the simple eqs. 12
and 13 is possible. Moreover, {r,b,CV C4, BP, BS} are themain parameters an operator should consider to decide whether
to invest in SDH or Packet interface, independently of the
network topology and complexity.
We finally show the comparison between the SA heuristic
and the exact ILP approach in the optimization of the number
of interfaces for static traffic and a realistic network. The
comparison has been carried out for the two SDH and Packet
technologies and for both classes of service (protected andunprotected). The chosen network is the Pan-European Re-
search Network Geant2* [24] (shown in figure 2). The results
SE
IE
UKNL
PL
CZFR
BEDE
CHXX
AT
SI HR
HU
ES IT
GR
Fig. 2. Geant2* network topology.
are depicted in figure 3. The plotted graphs show that the
difference between the results obtained by SA and ILP is
always beyond 10%. The static traffic is built starting from
a matrix (taken from the project MUPBED [24]) representing
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings
-
8/2/2019 Carrier Ethernet Over SDH
6/6
1e+1
1e+2
1e+3
0 5 10 15 20 25 30
ILP - PACKET - NO PROTSA - PACK - NO PROTILP - SDH - NO PROTSA - SDH - NO PROTILP - PACK - PROTSA - PACK - PROTILP - SDH - PROTSA - SDH - PROT
NumberofDeploy
edInterfaces
Traffic Scale Factor
Fig. 3. Number of deployed interfaces in Geant2* network.
the estimation of actual traffic volumes between the nodes of
the network. The total traffic volume has been varied over a
limited range to ensure ILP convergence within a reasonable
computational time. ILP has been performed on a standard pc
using CPLEX software. As expected the computational time
for SA is on average much lower (a few hours) than for ILP
(a few days).
The interfaces deployed in the Packet network are always
less than those in the SDH one, not only because of the
granularity constraint which affects SDH circuits, but also
because the multipath effect is nearly absent. In fact at thehighest traffic scale factor, the average requested bandwidth is
925 Mbit/s, which in most cases is too low to conveniently
exploit VCAT multiple paths. Moreover, the Geant2* topology
is characterized by a low average nodal degree, thus making
few different paths available for VCAT.
VII. CONCLUDING REMARKS
This paper deals with the problem of optimizing networks
able to support Carrier Ethernet connections. The purpose is
to evaluate the advantages of flexible access to bandwidth,
deriving from the adoption of layer-2 switching technologies
(such as PBB-TE and T-MPLS), against the disadvantages of
the fixed bandwidth granularity of SDH networks. We have
presented an integer linear programming formulation of the
problem of minimizing the number of interfaces deployed in
the network to setup a given static connection-request matrix,
and proposed a less computationally-hard heuristic based on
simulated annealing.
This preliminary study allowed us to identify a set of
fundamental parameters to evaluate the pros and cons of mi-
grating from legacy technologies to newfangled ones. It should
be finally commented that our study leads to comparisons
based solely on the number of interfaces, but a 10 Gigabit
Ethernet interface is reputed to be less expensive than an SDH
interface. Hence, the advantages of the Packet in terms of
monetary CAPEX value are probably more remarkable than
what predicted here.
REFERENCES
[1] R. Santitoro, Metro Ethernet Services - a technical overview, MetroEthernet Forum, 2003.
[2] IEEE P802.1Qay, IEEE Draft Standard for Local and MetropolitanArea NetworksVirtual Bridged Local Area Networks - Amendment:Provider Backbone Bridge Traffic Engineering, 2009.
[3] ITU-T Standard G.8110.1/Y.1730.1, Architecture of Transport MPLS(T-MPLS) layer network, 2006.
[4] Lucent-Technologies, Carrier Ethernet - defined, 2005.[5] TPACK, PBT: Carrier Grade Ethernet transport, 2006.[6] TPACK, Transport MPLS - a new route to Carrier Ethernet, 2006.[7] D. Allan, N. Bragg, and A. McGuire, Ethernet as Carrier Transport
Infrastructure, IEEE Communications Magazine, vol. 44, no. 2, pp. 134 140, 2006.
[8] A. Schmid-Egger and A. Kirstadter, Ethernet in core networks: Atechnical and economical analysis, in HPSR 2006 Workshop, Poznan,Poland, 2006, pp. 135 140.
[9] A. Kirstadter, C. Gruber, J. Riedl, and T. Bauschert, Carrier-GradeEthernet for packet core networks, in Proceedings of SPIE - The
International Society for Optical Engineering, vol. 6354 I, Gwangju,Korea, Republic of, 2006.[10] M. Batayneh, D. A. Schupke, M. Hoffmann, A. Kirstadter, and
B. Mukherjee, Optical network design for a Multiline-Rate Carrier-Grade Ethernet under transmission-range constraints, Journal of Light-wave Technology, vol. 26, no. 1, pp. 121 130, 2008.
[11] M. Batayneh, D. A. Schupke, M. Hoffmann, A. Kirstadter, andB. Mukherje, Carrier-Grade Ethernet over WDM under maximumtransmission range (TR) constraints of signal, in OFC/NFOEC 2008,San Diego, CA, United states, 2008.
[12] M. Batayneh, D. A. Schupke, M. Hoffmann, A. Kirstadter, andB. Mukherjee, Cost-efficient routing in Mixed-Line-rate (MLR) opticalnetworks for Carrier-Grade Ethernet, in OFC/NFOEC 2008, San Diego,CA, United states, 2008.
[13] R. Fu, Y. Wang, and M. Berger, Carrier Ethernet network control planebased on the Next Generation Network, in Proceedings of the 1st ITU-T Kaleidoscope Academic Conference, Innovations in NGN, K-INGN,
Geneva, Switzerland, 2008.[14] A. Atie, Leveraging SDH infrastructure to deliver Ethernet Services,
in IEEE LCW 2008, Beirut, Lebanon, 2008, pp. 51 57.[15] R. Sanchez and L. Raptis, Ethernet as a Carrier Grade technology: De-
velopments and innovations, IEEE Communications Magazine, vol. 46,no. 9, pp. 88 94, 2008.
[16] A. Reid, P. Willis, I. Hawkins, and C. Bilton, Carrier Ethernet, IEEECommunications Magazine, vol. 46, no. 9, pp. 96 103, 2008.
[17] R. Vaishampayan, A. Gumaste, S. Rana, and N. Ghani, Applicationdriven comparison of T-MPLS/MPLS-TP and PBB-TE - driver choicesfor Carrier Ethernet, in INFOCOM Workshops 2009, IEEE, April 2009,pp. 16.
[18] T. Engel, A. Autenrieth, and J.-C. Bischoff, Packet layer topologies ofcost optimized transport networks multi-layer netwok optimization, inONDM 2009, Feb. 2009, pp. 17.
[19] ITU-T Standard G.709/Y.1331, Interfaces for the Optical Transport
Network (OTN), 2003.[20] R. Guerin, H. Ahmadi, and M. Naghshineh, Equivalent capacity andits application to bandwidth allocation in high-speed networks, IEEE
Journal on selected areas in Communications, vol. 9, no. 7, pp. 968 981, 1991.
[21] D. McDysan, QOS & traffic management in IP & ATM networks.McGraw Hill, 2000.
[22] M. R. Garey and D. S. Johnson, Computers and Intractability: A Guideto the Theory of NP-Completeness. W.H. Freeman and Company, 1979.
[23] S. Kirkpatrick, C. D. Gerlatt, and M. P. Vecchi, Optimization bySimulated Annealing, Science, vol. 220, pp. 671680, 1983.
[24] MUPBED - Multi-Partner European Test Beds for Research Networking,Deliverable D1.2 - Revision of the reference architecture according tothe results of the project studies, 2005.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings