3526 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 72, NO.

6, JUNE 2024

Crosstalk-Aware Virtual Network Mapping in

Space-Division-Multiplexing Optical
Data Center Networks
Bowen Chen , Member, IEEE, Wenwen Zheng, Jun Yu, Danyang Zheng , Member, IEEE, Hong Chen,
Mingyi Gao , Weiguo Ju, Pin-Han Ho , Fellow, IEEE, Jason P. Jue, Senior Member, IEEE,
and Gangxiang Shen , Senior Member, IEEE

Abstract— This paper addresses the virtual network (VN) region has posed great challenges in network control and
mapping problems for the network profit optimization in management, particularly due to the rapid expansion of busi-
space-division-multiplexing optical data center networks (SDM- ness/enterprise activities that have imposed ever increasingly
ODCNs). We first define both link resource availability (LRA)
and node resource availability (NRA) for the VN mapping demands on the Internet services and transport capacity.
optimization, by which an integer linear program (ILP) model Another critical barrier of the bandwidth and computing
and two VN mapping approaches are proposed to achieve the resource expansion is the difficulty in the deployment of new
high network profit. Simulation results verify that our proposed network infrastructure of a similar scale around the world,
LRA VN mapping approach achieves greatly close network which leaves Internet service providers stuck in a “rigid”
performance to that by solving solutions of the integer linear
program model and significantly outperforms its counterpart situation. As a remedy, network virtualization techniques [1]
approaches. The improved network profits out of the VN map- have been used for resource/capacity slicing with quality of
ping is as a result of well suppressed average crosstalk, rejection service.
ratio of VNs, and spectrum fragmentation ratio in SDM-ODCNs. The previous studies only investigated the virtual nodes and
Index Terms— Space-division-multiplexing optical networks, virtual links that were separately mapped to the physical nodes
virtual network mapping, data center networks. and physical links for each virtual network (VN) [2], [3],
respectively. However, they did not consider the effect of the
I. I NTRODUCTION bandwidth requirements of virtual links and the computing
resource of virtual nodes on the VN mapping. Meanwhile,
T HE rapid growth of cloud-edge computing in the past
decade has served as a driving force for numerous
emerging Internet application such as network videos, data
the previous studies also did not consider the effect of the
bandwidth resources provisioning of physical links and the
center, and Internet of things (IoT). Nonetheless, the limited computing resources of physical nodes on the VN mapping.
bandwidth resources and computing resources at the edge Different from previous studies about virtual network mapping
problems, the novelty of this paper is described as follows.
Manuscript received 28 December 2022; revised 2 September 2023 We define two novel parameters that are the link resource
and 16 December 2023; accepted 3 February 2024. Date of publica-
tion 9 February 2024; date of current version 18 June 2024. This work availability (LRA) and the node resource availability (NRA)
was supported by National Key Research and Development Program of for achieving effective VN mapping. Both LRA and NRA are
China (2022YFB2903303), Natural Science Foundation of Jiangsu Province designed to better achieve the network performance in solving
(BK20200099), China Postdoctoral Science Foundation (2020T130460,
2020M671598), Jiangsu Postdoctoral Science Foundation (2020Z109), and the VN mapping problem. The LRA VN mapping approach
Jiangsu Engineering Research Center of Novel Fiber Technology and emphasizes the way that the virtual link with the highest value
Communication Network, Suzhou Key Laboratory of Advanced Optical of virtual link resource availability is preferentially mapped
Communication Network Technology. An earlier version of this paper
was presented at the 2021 Proceedings of ICOCN, Qufu, China [DOI: to the physical link with the highest value of physical link
10.1109/ICOCN53177.2021.9563685]. The associate editor coordinating the resource availability for each VN. Furthermore, the NRA
review of this article and approving it for publication was W. Shieh. VN mapping approach achieves by using the way that the
(Corresponding author: Bowen Chen.)
Bowen Chen, Wenwen Zheng, Jun Yu, Danyang Zheng, Hong Chen, virtual node with the highest value of virtual node resource
Mingyi Gao, and Gangxiang Shen are with the School of Electronic and availability are preferentially mapped to the physical node
Information Engineering, Soochow University, Suzhou, Jiangsu 215006, China with the highest value of physical node resource availability.
(e-mail: [email protected]).
Weiguo Ju is with the Institute of ICT Technology, China Information We propose the integer linear programming (ILP) model and
Consulting and Designing Institute Company Ltd., Nanjing, Jiangsu 210019, the LRA and NRA VN mapping approaches to achieve high
China. network profit and to reduce rejection ratio of VNs compared
Pin-Han Ho is with the Department of Electrical and Computer Engineering,
University of Waterloo, Waterloo, ON N2L 3G1, Canada. to the random (RAN) VN mapping approach in space-division-
Jason P. Jue is with the Erik Jonsson School of Engineering and Computer multiplexing optical data center networks (SDM-ODCNs).
Science, The University of Texas at Dallas, Richardson, TX 75080 USA. As shown in Fig. 1, with the SDM optical data center
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TCOMM.2024.3364359. network architecture, edge data centers (EDCs) are located
Digital Object Identifier 10.1109/TCOMM.2024.3364359 between the centralized data center (CDC) and terminal equip-
0090-6778 © 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 CHEN et al.: CROSSTALK-AWARE VN MAPPING IN SDM-ODCNs 3527

fragmentation and enhance the network availability [12], [13].

Considering the resource efficiency, the VN mapping strategy
was proposed to greatly improve the acceptance ratio and
the network revenue [14]. A self-organizing VN mapping
approach was built on a machine learning model that is trained
by using previous successful virtual node mapping information
to improve the network profit and acceptance ratio [15].
A spectrum-efficient VON mapping algorithm was proposed to
suppress inter-core XT and reduce spectrum occupancy [16],
but this study only proposed the heuristic VON mapping
algorithm and did not formulate the ILP model to optimize
the VONs mapping in SDM-ODCNs.
Fig. 1. A schematic of SDM optical data center network.
B. VN Mapping for SDM
The introduction of spatial domains in optical communi-
ment via optical fibers and provide the limited data processing cations has recently attracted considerable attention due to
capacity to the nearby end clients. The EDCs connect different the capacity depletion for the traditional wavelength division
terminal devices. i.e., different applications. The core network multiplexing (WDM) networks. In spite of the increased
is a SDM-ODCN, where end clients could launch intensive system capacity, SDM brings up new challenges when being
computing requests to either a nearby edge data center or the deployed in the optical data center network (ODCN) [17].
centralized data center for processing. For example, the spectrum fragmentation will seriously affect
The organization of the rest paper is as follows. The network resource efficiency. Meanwhile, XT is another impor-
related work and our contributions are described in Section II. tant issue in order to suppress crosstalk happening between
The system model and problem statement are discussed in two adjacent cores in SDM optical networks. Thus, an XT-
Section III. Section IV presents the cost and revenue model, aware spectrum defragmentation algorithm was proposed to
and evaluation metrics. The resource availability and inter-core achieve low defragmentation and to improve the spectrum
XT are described in Section V. Sections VI and VII pro- efficiency [18]. A spectrum defragmentation algorithm based
pose the ILP model and heuristic VN mapping approaches, on dedicated path protection was proposed to address the
respectively. Section VIII shows the simulation and results.
spectrum efficiency problems in SDM-EONs [19]. To further
Section IX is the complexity analysis. Section X concludes
consider that fact that inter-XT may occur between adja-
the paper.
cent cores in optical networks with MCFs, [9] proposed
an “on-demand” spectrum and core allocation to suppress
II. R ELATED W ORK AND O UR C ONTRIBUTIONS XT and to greatly reduce spectrum fragmentation. However,
A. VN Mapping for DC these proposed XT-aware algorithms mainly investigated the
dynamic traffic demands rather than the static scenario in
Different optical switching nodes are connected by using
SDM-EONs.
multicore fiber in the SDM-ODCN. By employing multi-core
fibers (MCF), an optical data center network can achieve much
higher communication capacity than that with single-core fiber C. VN Mapping for Network Profit
(SCF) [4]. Such advantage, nonetheless is at the expense of The network profit was generally defined as the network rev-
increased inter-core crosstalk (XT) that may impose non-trivial enue minus the network cost. In [20], a hybrid VN embedding
impairment on optical transmission quality by suppressing heuristic algorithm was developed to solve the problems of
XT. Through network virtualization, SDM-ODCNs can meet bandwidth efficiency, where a baseline policy, named heuristic
sophisticated service requirements by breaking the binding VN embedding algorithm, was introduced for comparison.
relationship between the physical devices and logical service In elastic optical networks, [21] proposed the survivable VON
layers [5], [6], [7], where core selection, inter-XT mitigation, mapping algorithms to decrease the network cost. For each
and spectrum resource allocation can be jointly considered for virtual link in the mapping process, authors in [22] employed
achieving the desired service requirements [8], [9]. the dedicated-path protection to ensure high survivability in
Current research efforts on VNs mapping predominantly the optical networks. In order to improve the network profit,
focus on improving bandwidth efficiency, network cost, and backup cost, and resource usage, authors in [23] proposed
energy efficiency. The VNs mapping problem involves making the VN mapping with the protection way against link and
decisions on how to map the optical data center network by node failures. A framework was presented to reduce the
using heuristic mapping approaches. For example, a novel network cost and improve the quality of services and the
energy-efficient embedding method was proposed to map resource utilization [24]. Although VON mapping approaches
heterogeneous VNs onto DC infrastructure to improve the with the link importance degree and the node importance
bandwidth efficiency and to reduce energy consumption [10], degree were proposed to improve the network profit in SDM-
[11]. For making efficient use of spectrum resources, a hybrid EODCNs [25], this study did not consider the effect of XT on
VN mapping approach was proposed to decrease spectrum VONs mapping. These above existing studies only investigated

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 3528 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 72, NO. 6, JUNE 2024

to reduce the network cost of VNs mapping rather than

considering the effect of the link resource availability and
the node resource availability on the network profit in SDM-
ODCN.

D. Our Contributions
Our contributions are described as follows:
1) In order to obtain the high network profits, we for-
mulated link resource availability and node resource
availability to optimize VN mapping in SDM-ODCNs.
2) We proposed the ILP model to obtain the maximum net- Fig. 2. (a) A fiber link with seven cores and (b) XT between two adjacent
cores.
work profit by reducing the network cost while mapping
a given set of VNs.
3) We also proposed the LRA and NRA VN mapping
approaches to improve the network profit. We can verify
that the simulation results of the LRA VN mapping
approach are greatly close the optimal solutions of the
ILP model.
4) We can find that the LRA VN mapping approach not
only improves the network profit, but also reduces the
average crosstalk and the rejection ratio of VNs com-
pared to the NRA and RAN VN mapping approaches.

III. S YSTEM M ODEL AND P ROBLEM S TATEMENT

A. Network Model
In the VN mapping, SDM-ODCN is defined as an undi-
rected graph, GP = {V P , E P , O P , F P , C P }, in Fig. 1. Fig. 3. An example of a VN mapping to SDM-EODCN.
SDM-ODCN also calls the physical network (PN). V P =

vPi |i = 1, 2, · · ·, |V P | represents a set of optical switching
nodes, E P = eiP |i= 1, 2, · · · , |E P | denotes set of fiber
links, OP = oiP |i = 1, 2, · · · , |O P | is a set of fiber cores, E nV = {env i |i= 1, 2, · · · , |E nV |} be a set of virtual links (VLs),
F P = fPi |i = 1, 2, · · · , |F P | is a set of frequency slots cnv i in C nV = {cnv i |i= 1, 2, · · · , |V nV |} be the computing
(FSs) on a fiber core, and C P = ciP |i = 1, 2, · · · , |V P | resource requirements of a virtual node vvni , and bnv i in
denotes the available computing resources of a data center. B nV = {bnv i |i= 1, 2, · · · , |E nV |} be the bandwidth resource
|V P | represents the number of optical switching nodes, |E P | requirements of a virtual link env i , where |V nV | represents the
denotes the number of fiber links in a SDM-ODCN, |O P | number of virtual nodes and |E nV | represents the number of
is the number of cores in each MCF, and |F P | denotes the virtual links on the nth VN.
number of available FSs in a fiber core. The optical switching Virtual node represents virtual DCs and virtual links are the
node connected to each DC in SDM-ODCN includes optical bandwidth resource requirements that are connected by the
transponders and optical regenerators. The optical switches virtual links for each VN. A physical node on SDM-ODCN
consist of a sets of wavelength cross-connects (WXCs) [8]. includes the data center and the optical switch, which provides
Optical amplifiers are placed before and after the WXC to the computing resources. The physical link on SDM-ODCN
compensate fiber and node losses. Unlike previous work, each denotes the fiber link that provides the spectrum resources. The
fiber link contains seven cores with a corresponding core computing resources provided by the physical node are not less
number as shown in Fig. 2 (a). In this paper, the FSs of each than the required computing resources by virtual nodes, the
core and the computing resources available at each DC are virtual nodes can be mapped to the physical nodes. Similarly,
considered resources to be allocated. In Fig. 2 (b), the adjacent when the spectrum resources of physical links (PLs), i.e., fiber
cores of the selected core 0 are the cores 1, 5, and 6. XT occurs links, are not less than the spectrum resource requirements of
between two adjacent cores [26]. virtual links, we map the virtual links to the physical links.
Fig. 3 shows an example of a VN mapping to SDM-
EODCN. A VN composed of virtual nodes, a, b, and c,
B. VN Mapping Model and virtual links, a-b, b-c, and a-c. The required computing
In VN mapping model, let a set of VNs GV = resources of virtual nodes are 7, 5, and 8 units on virtual nodes,
{gvn |n = 1, 2, · · · , |GV |} be mapped to a physical network, a, b, and c. The bandwidth resource requirements are 40 Gbps,
SDM-EODCN, denoted as GP = {V P , E P , O P , F P , C P }, 10 Gbps, and 100 Gbps on virtual links, a-b, b-c, and a-c,
where gvn = {V nV , E nV , C nV , B nV } denotes the nth VN. Let respectively. The bandwidth resources provisioning are enough
V nV = {vvni |i= 1, 2, · · · , |V nV |} denote a set of virtual nodes, on each fiber link in SDM-EODCN. The computing resources

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 CHEN et al.: CROSSTALK-AWARE VN MAPPING IN SDM-ODCNs 3529

provisioning are 68, 75, 42, and 57 units on the data centers determined by the number of successfully mapping VNs in
that are located at the optical switching nodes, A, B, C, and D, SDM-ODCNs. The network revenue of mapping each VN
respectively. We can map the virtual nodes a, b, and c to the is decreased due to the spent computing resources and the
optical switching nodes A, C, and B, respectively. The virtual required bandwidth resources, where the more the computing
links, a-b, b-c, and a-c are mapped to the fiber links A-B, B-C, and bandwidth resources are spent, the higher the network
and A-C, respectively. cost is, leading to the reduced network revenue. The network
revenue model of the VN mapping is defined as follows.
X X X X
C. Problem Statement RGV = cnv i + θ × bnv i (2)
Our objective attempts to maximize the network profit in n∈GV i∈V n
V n∈GV l∈E n
V

the VN mapping process. We are given a physical network, where GV , V nV , and E nV represent a set of VNs, a set of
Gp = {Vp , Ep , Op , Fp , Cp }, a set of VNs, Gv = {gvn |n = virtual nodes, and a set of virtual links on the nth VN,
1, 2, . . . , |Gv |}, the unit cost of the optical transponder and the respectively. cnv i and bnv i represent the required computing
optical regenerator, which denotes as TC and RC, respectively. resources of the ith virtual node and the required bandwidth
To form a VN mapping in SDM-ODCNs, we firstly need to resources of a virtual link l∈E nV on the nth VN, respectively.
find physical nodes to be mapped by virtual nodes, where the θ represents the revenue equilibrium factor of the virtual nodes
computing resources of the virtual nodes are met on a VN. and links.
Next, a set of virtual links of the VN is mapped to a physical
network where the spectrum continuity, spectrum consecu-
tiveness, and inter-core XT threshold constraints are met. B. Evaluation Metrics
The spectrum consecutiveness constraint is that the allocated In order to fairly verify the network performance of the
spectrum must be chosen from contiguous frequency slots in proposed VN mapping approaches, including the ILP model
the frequency domain on a fiber link. The spectrum continuity and the heuristics, a number of metrics is employed as follows.
constraint is presented as the allocation of the same spectrum 1) Total network profit (T P GV ): obtained by subtracting
on each link along a path. Finally, the optical transponders the cost CGV from the benefit RGV of the VN mapping:
should be configured at two optical switching nodes of each T PGV = RGV − CGV (3)
physical link, while optical regenerators should be configured
′


in any intermediate switching node in the case that the optical 2) Rejection ratio (RrGV ): the ratio of failed VNs GV
signal exceeds the optical reachability of the transmission path. mapping to the total number of requested VNs (|GV |):
Motivated by the need of an effective approach for solving the
G′V
inter-core XT-aware VN mapping problems in SDM-ODCNs, RrGV = (4)
we formulate an ILP model and propose two heuristic VN |GV |
mapping approaches with the link resource availability (LRA) 3) Spectrum fragmentation ratio (SF RGP ): referred to the
and node resource availability (NRA) to improve the network ratio of the total spectrum fragmentations to the total number
profit and spectrum efficiency in SDM-ODCNs. of frequency slots: P
F Sl
l∈E P
IV. C OST AND R EVENUE M ODEL , AND E VALUATION SF RGP = (5)
F St
M ETRICS
where F S l and F S t are the number of spectrum fragmenta-
A. Cost and Revenue Model tions of a fiber link, l∈E P , and the total number of frequency
In SDM-ODCNs, the network cost of a VN mapping slots.
depends on the number of optical transponders and optical
regenerators configured to support the required interconnection V. R ESOURCE AVAILABILITY AND I NTER -C ORE
and bandwidth. Two optical transponders of a certain link rate C ROSSTALK
are firstly required for a virtual link mapping, and secondly the
In order to efficiently map VNs to a SDM-ODCN, we first
number of optical regenerators is configured on the interme-
propose the link resource availability (LRA) and node resource
diate switching nodes when the distance of the transmission
availability (NRA) to effectively implement VN mapping.
path exceeds maximum reachability of an optical signal. This
We also discuss the inter-core crosstalk when mapping a given
leads to the network cost of VN mapping as follows:
set of VNs.
X X X X
CGV = 2 × T C × Nln + RC × Mln
n∈GV l∈E n n∈GV l∈E n
A. Link Resource Availability
V V

(1) Here, we define the link resource availability as the import

parameter for LRA VN mapping. The link resource availabil-
where Nln and Mln represent the number of optical transpon- ity includes the physical link resource availability (PLRA)
ders and optical regenerators for a virtual link l∈E nV , and virtual link resource availability (VLRA) for achieving
respectively. effective VN mapping process on the physical layer and
The network revenue of VN mapping with limited com- virtual layer, respectively. Firstly, we define the physical link
puting resources and bandwidth resources is exclusively resource availability as the amount of computing resources

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 3530 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 72, NO. 6, JUNE 2024

provided by data center nodes, available frequency slots of

the adjacent physical links, and distance of physical links.
With the spectrum adjacency of each core in each fiber link in
SDM-ODCN, the PLRA of physical link e ∈ E P of the two
terminating node s and d is measured as follows:
1 sa (e)
P LRAe∈E P = × csP + cdP ×


2 |C|−1
P PP | e
|F Fig. 4. (a) An example of a VN and (b) the VLRA values on virtual links.
ϑfk + 1
k=0 fk
Dmax − D(s,d)
× (6)
Dmax − Dmin
e
|C|−1 Tke Tk,i
X XX
e


sa (e) = Tk,i −j+1 (7)
k=0 i=1 j=1

where csP and cdP represents the provided computing resources

from a physical source node s to a physical destination node d,
respectively. Dmax and Dmin is the maximum and minimum
distance of the transmission paths for each node-pair of SDM-
ODCN, respectively. D(s,d) is the distance of a shortest path
Fig. 5. (a) An example of a physical network and (b) the auxiliary graph.
from a physical node s to a physical node d on the eth physical
link. ϑefk represents the usage of the fk th FS on the kth core
of a physical link e∈E P , where its value is 1 if the fk th FS is
link resource availability for all virtual link of Fig. 5 (a), which
free and it’s value is 0, otherwise. sa (e) denotes the spectrum
are calculated by Equation (7).
adjacency on the physical link e∈E P in formula (7), Tke and
e In order to simplify the VN mapping process, we can
Tk,i represent the number of the ith spectrum fragmentations
construct auxiliary graph (AG) by configuring all the paths
and the number of FSs on the kth core of a physical link
based on all the node pairs on the SDM-EODCN in advance.
e∈E P , respectively. |C| and |F P | denote the number of cores
By constructing auxiliary graph, we can easy map virtual links
of each fiber link and total FSs for each fiber core, respectively.
and virtual nodes to the physical path and the physical nodes,
Similarly, we also propose the virtual link resource avail-
respectively. The reason is that we do not need to calculate a
ability (VLRA), which is written as follows:
path for mapping a virtual link on the SDM-EODCN due to
1 the preconfigured path on auxiliary graph. An auxiliary graph
V LRAi∈E nV = × bnv i × csv + cdv


(8)
2 of the physical network is constructed for Fig 5.
where bnv i represents the bandwidth resource requirements of a Fig. 5 (a) shows a topology with nodes 1, 2, 3, and
virtual link i ∈ EnV on the nth VN, and csv and cdv represent the 4 representing a physical network with four physical nodes.
computing resource requirements located at a source node s The circles near the physical nodes represent the data cen-
and a destination node d for a virtual link i∈E nV , respectively. ters deployed on those physical nodes. On each data center,
Both PLRA and VLRA not only consider the spectrum the number on each circle denotes the available computing
resources and the distance of physical links, but also consider resources. On each physical link, the number represents the
the computing resources on physical nodes. Both PLRA and transmission distance between physical nodes. The grid on the
VLPA are aimed at minimizing the mapping network cost physical link represents the occupied status of frequency slots.
when mapping a given set of VNs. Obviously, the PLRA value First, Dijkstra’s algorithm is employed to calculate the shortest
depends on the provided computing resources by the physical paths based on all physical node-pairs. The transmission dis-
nodes, the spectrum usage of all adjacent physical links, and tance of all physical node-pairs on the auxiliary graph and the
the distance of physical links. VLRA depends on the required spectrum resources are obtained along this path in Fig. 5 (b).
bandwidth and computing resources of virtual links and nodes. The principle of spectrum consecutiveness and continuity
We can employ both PLRA and VLRA to map the VNs to the needs to be satisfied when searching for the available FSs for
SDM-ODCN. A reasonable resource allocation can increase each physical link in Fig. 5 (a). For example, the path of node-
the number of successful VN mapping and improve network pair (1, 3) is 1-2-3, so its physical link distance is 1600 km,
profit. Obviously, the appropriate VN mapping approaches can the number of available FSs is obtained by cross-referencing
also reduce the VN mapping cost. free FSs on fiber links 1-2 and 2-3. Finally, the PLRA between
Fig. 4 shows an example of the VLRA values for VN all physical node-pairs is calculated by using Equation (6) in
mapping. In Fig. 4 (a), there is a virtual network consisting Fig. 6 (a). The mapping result is shown in Fig. 6 (b).
three virtual nodes and links. Three virtual nodes, A, B,
and C require computing resources with 3, 2, and 6 units,
respectively, and three virtual links, A-B, B-C, and A-C B. Node Resource Availability
require bandwidth resources 395 Gbps, 270 Gbps, and A-C Here, we define the node resource availability as the import
222 Gbps, respectively. Fig. 4 (b) shows the values of virtual parameter for achieving NRA VN mapping. For the node

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 CHEN et al.: CROSSTALK-AWARE VN MAPPING IN SDM-ODCNs 3531

Fig. 7. The VNRA values on a virtual network.

Fig. 6. (a) An auxiliary graph of VN mapping with PLRA and (b) the VN
mapping result with PLRA.

resource availability, we define it as physical node resource

availability (PNRA) and virtual node resource availability
(VNRA) on the physical layer and virtual layer, respectively.
We propose the physical node resource availability (PNRA)
to achieve effective VN mapping problems. PNRA is defined
as the available spectrum and computing resources and the
data-center nodes connected on a PL, respectively, which is
written as follows (9): Fig. 8. (a) An auxiliary graph of VN mapping with PNRA and (b) the VN
1 X sa (e) mapping result with PNRA.
P N RAs∈V P = csP × × (9)
|E P | |C|−1
P PP | e
|F
e∈E P
ϑfk + 1
k=0 fk which shows in the diamond beside each optical switching
where P N RAs∈V P is the PNRA of the sth physical node, V P node in Fig. 8 (a), is computed according to formula (8). The
and EP are a set of PNs and a set of PLs on SDM-ODCN, VN in Fig. 4 (a) is successfully mapped to the auxiliary graph
respectively. csP denotes the provided computing resources on of the SDM-ODCN in Fig. 8 (a). Obviously, three virtual nodes
a PN, s∈V P . ϑefk represents the usage of the fk th FS on the A (3 units), B (2 units), and C (6 units) are mapped to three
kth core of a PL e∈E P , where its value is 1 if the fk th FS physical nodes 4 (50 units), 1 (40 units), and 2 (45 units),
is free, otherwise, its value is 0. sa (e) represents spectrum on auxiliary graph, respectively. Three virtual links A-B (395
adjacency on a physical link, e∈E P , which is obtained by Gbps), B-C (270 Gbps), and A-C (222 Gbps), are mapped to
using Equation (7). Tke represents the number of spectrum three physical links 1-4, 4-2, and 1-2 by using both PNRA and
fragmentations on the kth core of a physical link e∈E P . Tk,i e VNRA, respectively. The mapping result is shown in Fig. 8 (b).
represents the number of FSs of the ith spectrum fragmentation
on the kth core of a physical link e∈E P . |E P | denotes the C. Inter-Core Crosstalk of MCF
number of adjacent physical links. |C| represents fiber cores We mainly address the problems of core and spectrum
of a fiber link, and |F P | is total FSs of each fiber core. assignment in MCF-based SDM-ODCN. XT occurs between
Similarly, we can also propose the virtual node resource two adjacent fiber cores when the overlapping frequency
availability (VNRA). The VNRA is defined as the require- slots are assigned. We consider that FSs are available if XT
ments of computing resources for virtual nodes and the of the selected FS does not exceeds the threshold XTmax in
requirements of bandwidth resources for virtual links that are SDM-ODCN.
directly connected to virtual nodes, which is written as follows: The resource allocation process in SDM-ODCN must con-
M sider the physical impairment of the transmitted signal caused
1 X
V N RAj∈V n = cnv j + × bnj,m (10) by XT between adjacent fiber cores during the transmission
V
M m=1 v
process. XT in Fig. 2(b) occurs when optical signals are
where V N RAj∈V nV denotes VNRA of the jth virtual node transmitted on the same FSs between two adjacent cores.
of the nth VN, V nV represents a set of virtual nodes of When the XT value is too large, the optical signal will be
n n
the nth VN, cv j and bv j,m represent the computing resource degraded, and the transmission quality of the optical signal
requirements and bandwidth resource requirements on the mth cannot be guaranteed. Therefore, a XT threshold needs to be
VL for the jth virtual node on the nth VN, respectively, and introduced to guarantee the quality of service, which is defined
M represents the number of VLs that are directly adjacent to as the XT constraint. A VN mapping will block when the XT
virtual nodes. Fig. 7 is the VNRA values of a virtual network value of the optical channel is greater than the XT threshold.
as shown in Fig. 4 (a) and the number in the red box is the Optical signals of different modulation formats can tolerate
virtual node resource availability. different XT thresholds.
To simplify the VN mapping process, we should construct The XT influence depends on the core pitch between two
the auxiliary graph of VN mapping. The constructed auxiliary fiber cores. The smaller the distance is, the more serious the
graph of PNRA is similar to the auxiliary graph of PLRA crosstalk is. In addition, the XT influence is also related to the
in Fig. 5. After all shortest paths are preconfigured for all number of adjacent fiber cores, the length of the optical fiber,
physical node-pairs, the PNRA value of each physical node, and the physical parameters of the optical fiber, and so on.
Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 3532 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 72, NO. 6, JUNE 2024

In order to calculate XT of SDM-ODCN, the basic calculation by XT of the adjacent cores. The XT value based on the
method of the XT value between two cores is introduced in spectrum state can avoid the influence of the overlap between
[27] and [28], which can be expressed by formula (11): the selected spectrum block and the occupied spectrum block
of adjacent cores.
n − n × e−2×(n+1)×h×L
XT = (11)
1 + n × e−2×(n+1)×h×L
VI. ILP M ODEL OF VN M APPING
where the notation δ is the inter-core crosstalk value between
2k2 r
two adjacent fiber cores, n, h = βω th
, and L represent the In order to optimize VN mapping, an ILP model is proposed
number of adjacent fiber cores, the average increase of the to minimize the network cost by VN mapping with the XT
XT value per unit length, and the length of a fiber link, constraint in SDM-ODCNs.
respectively. k and β are the physical parameters of the optical Input parameters:
fiber, which refer to the coupling coefficient, propagation GV : a given set of VNs.
constant, respectively. r and ωth are the bending radius and n : the nth VN, n ∈ GV .
center-to-center distance, respectively. E nV : a given set of VLs on the nth VN.
When evaluating the XT value, we must consider not only V nV : a given set of virtual nodes on the nth VN.
the effect of physical parameters in the fiber links, but also the E P : a given set of PLs on a SDM-ODCN.
effect of the spectrum occupancy state between two adjacent V P : a given set of PNs on a SDM-ODCN.
cores. Therefore, when searching for the bandwidth require- (i, j): a VL from a virtual nodes ito a virtual nodes j.
ments of each virtual link, we must introduce the inter-core (k, l): a PL from a PN k to a PN l.
spectrum state factor (IC-SSF) to evaluate the effect of the (s, d): a physical path from a PN s to PN d.
spectrum occupancy state between the adjacent cores on the F sni,j : the number of the required FSs for a VL (i, j) on
XT value of the currently pre-allocated spectrum resources. the nth VN.
The IC-SSF value is calculated by the following formula (12): n
Ni,j : the required number of the optical transponders on a
 X
rl
X VL(i, j) on the nth VN.


 τ1 × F j , if Fjrl > 1, j ∈ F Mi TC: the unit cost of an optical transponder.
Mfcli j = r∈n
X
rl
r∈n
X (12) RC: the unit cost of an optical regenerator.


 τ0 × F j , if Fjrl ≤ 1, j ∈ F Mi R: the maximum reachability of optical signal for an optical
r∈n r∈n
regenerator.
where Mfcli j
denotes the IC-SSF value of the adjacent cores cnv i : the computing resource requirements of the ith virtual
of the jth FS in the ith candidate spectrum block on the fiber node on the nth VN.
core c of the fiber link l. τ0 and τ1 are the weight adjustment csp : the number of available computing resources at the sth
factors of the IC-SSF value. r and n are an adjacent core physical node in the SDM-ODCN.
to the working core c and a set of adjacent fiber cores, |F P | : the maximum number of FSs on a core of a fiber
resepctively. Fjrl represents the occupancy status of the jth link.
FS on an adjacent fiber core c of the fiber link l. If P the jth FS D(k,l) : the distance of a PL (k, l) in the SDM-ODCN.
is occupied, then Fjrl = 1, otherwise, Fjrl = 0. r∈n Fj
rl C P : a set of fiber cores in a fiber link.
represents the total occupancy status of the jth FS on all F P : a set of FSs in a fiber core.
adjacent cores of the fiber link l. F M i is the ith candidate XT f,ck,l : the XT value generated by the f th FS of the cth
spectrum block, where its range of the searching spectrum core on a PL (k, l).
resources is from the index fi to the index fi + F S − 1, XTmax : the maximum XT tolerance of FSs on adjacent fiber
i.e., F M i = {fi , fi + 1, fi + 2, . . . , fi + F S − 1}. FS is the cores in SDM-ODCN.
bandwidth requirements of a virtual link. Λ: a very large positive number, which is larger than the
The crosstalk constraint must be satisfied by the XT thresh- total number of FSs on the whole SDM-ODCN.
old when searching and selecting available FSs on the fiber M: the total number of cores on a fiber link.
core. The same indexes of FSs on the adjacent fiber core Variables:
should be avoided as much as possible, so as to avoid the mni,j : the total number of the required optical regenerators
XT occurring. Therefore, based on the IC-SSF value between for a VL (i, j) for the nth VN.
the adjacent cores, we can obtain the XT value Acl fi between
xn(i,j),(s,d) : binary variable. Its value is 1 if a VL (i, j)
adjacent cores, which is expressed by formula (13): on the nth VN is mapped to a physical path (s, d) in the
SDM-ODCN, and its value is 0, otherwise.
n − n × e−2×(n+1)×h×L n
y(i,s) : binary variable. Its value is 1 if the ith virtual node
Acl
fi = 10 × lg
1 + n × e−2×(n+1)×h×L for the nth VN is mapped to the sth PN in the SDM-ODCN,
FS and its value is 0, otherwise.
× (13)
Mfcli j
P
FS + φn,f,c
(i,j),(k,l) : binary variable. Its value is 1 if and only if
j∈F Mi
the fth FS on the fiber core c of a PL (k, l) is occupied by a
where Aclfi represents the XT value based on the IC-SSF value, VL (i, j) of the nth VN, and its value is 0, otherwise.
which corresponds to P the fi th spectrum block on the cth fiber pn(i,j),(k,l) : binary variable. Its value is 1 if a PL (k, l) is
core of a fiber link l. j∈F M i Mfcli j refers to the value of the occupied by a VL (i, j) of the nth VN, and its value is 0,
fi th spectrum block on the cth core of a fiber link l affected otherwise.
Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 CHEN et al.: CROSSTALK-AWARE VN MAPPING IN SDM-ODCNs 3533

cn(i,j),c : binary variable. Its value is 1 if the cth core is Formula (23) not only ensures the selected physical nodes
occupied by a VL (i, j) on the nth VN, and its value is correspond a source node s to a destination node d of the
0 otherwise. physical path (s, d) for a virtual link (i, j), but also ensures
n that a virtual link (i, j) only occupies one physical link.
w(i,j),f : binary variable. Its value is 1 if the f th FS is
occupied by a VL (i, j) on the nth VN, and it’s value is 0, 3) Flow conservation constraint:
otherwise. X X
The ILP model is mathematically formulated as follows: xn(i,j),(k,l) × F sni,j − xn(i,j),(l,k) × F sni,j
l∈V P ∪V n l∈V P ∪V n
Objective:  P V

The objective of the ILP model is to maximize the network n

 F si,j , k = i

revenue for the VN mapping. It can be expressed as follows: = −F sni,j , k = j , ∀n ∈ GV , (i, j) ∈ E nV (24)
Maximize: 
0, otherwise

P RGV = RGV − CGV (14)
Formula (24) guarantees that the number of output flow, i.e.,
All VNs can be mapped to the SDM-ODCN for the ILP FSs, of the optical channel on each physical node except the
model, so all the VN mappings are accepted, and the network source node and the destination node is equal to that of the
profit is fixed. Therefore, our objective of optimizing the ILP input flow, i.e., FSs.
model of VN mapping can be translated into minimizing the 4) Spectrum adjacency constraint:
network cost for a given set of VNs as follows:
|F P |−1 |F P | w
Minimize: X X X
n
X X X X w(i,j),f ≥0 (25)
n
2 × T C × Ni,j + RC × mni,j (15) r=2 w=r+1 f =w−r
n∈GV l∈E n
V n∈GV l∈E n
V |F P |−1 |F P | w r
X X X X
n
Constraints: w(i,j),f ≤ t (26)
1) Virtual node mapping constraint: r=2 w=r+1 f =w−r t=1

Formulas (25) and (26) ensure that the number of consecu-

X
n
y(i,s) = 1, ∀i ∈ V nV (16)
s∈V P tively occupied FSs on each fiber link in SDM-ODCN does not
X
n exceed the number of total FSs, which satisfies the spectrum
y(i,s) ≤ 1, ∀s ∈ V P (17)
n
adjacency constraint.
i∈V V
5) Spectrum capacity constraint:
Formulas (16) and (17) guarantee that each virtual node on X X X n,f,c
the nth VN is mapped to a unique physical node. φ(i,j),(k,l)
X n∈GV (i,j)∈E n
V
f ∈F P
n
y(i,s) csp ≥ cnv i , ∀i ∈ V nV (18)
≤ |F P | , ∀c ∈ C P , (k, l) ∈ E P (27)
s∈V P X
cn(i,j),c × F sn(i,j) ≤ φn,f,c
(i,j),(k,l) ,
X
n
y(i,s) cnv i ≤ csp , ∀s ∈ V P (19)
f ∈F P ,(k,l)∈E P
i∈V n
V
∀n ∈ GV , (i, j) ∈ E nV , c ∈ C P (28)
Formulas (18) and (19) ensure that the computing resources
of the physical node meet the required computing resources cn(i,j),c ≥ φn,f,c
(i,j),(k,l) , ∀n ∈ GV , (i, j) ∈ E nV ,
by the mapped virtual nodes. c ∈ C P , f ∈ F P , (k, l) ∈ E P (29)
2) Virtual link mapping constraint: X
X pn(i,j),(k,l) × F sni,j = φn,f,c
(i,j),(k,l) ,
n
w(i,j),f = F sni,j , ∀n ∈ GV , (i, j) ∈ E nV (20) f ∈F P ,c∈C P
f ∈F P
  X ∀n ∈ GV , (i, j) ∈ E nV , (k, l) ∈ E P
n
w(i,j),f − n
w(i,j),f − 1 × (−Λ) ≥ n
w(i,j),z (30)
+1
z∈[f +2,|FP |] n
w(i,j),f ≥ φn,f,c n
(i,j),(k,l) , ∀n ∈ GV , (i, j) ∈ E V ,
(21) c ∈ C P , f ∈ F P , (k, l) ∈ E P (31)
  X
n
w(i,j),f − 1 × Λ + F sni,j ≤ n
w(i,j),f ,
Formula (27) guarantees that the total number of FSs on a
f ∈|F p |
PL does not exceed the total spectrum capacity of a physical
∀n ∈ GV , (i, j) ∈ E nV , f ∈ F P (22) link. In order to decrease the computational complexity of the
Formulas (20) to (22) ensure spectrum consecutiveness and formula (27), we can reduce the dimension of the computation
continuity constraints on the lightpath for each virtual link space of a variable, φn,f,c (i,j),(k,l) . We introduce three variables,
mapping. cn(i,j),c , w(i,j),f
n
, and pn(i,j),(k,l) , to limit the dimension of
X X computation space of a variable, φn,f,c (i,j),(k,l) . Thus, the formulas
xn(i,j),(s,d) − xn(i,j),(d,s) = y(i,s)
n n
− y(j,s) ,
from (28) to (31) are used to decrease the dimension space of
d∈V P d∈V P
a binary variable, φn,f,c(i,j),(k,l) , in order to reduce the computa-
∀n ∈ GV , (i, j) ∈ E nV , s ∈ V P (23) tional complexity.

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 3534 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 72, NO. 6, JUNE 2024

6) Core uniqueness constraint: is the total virtual links of the nth virtual network. Then
X we take out the virtual link with the highest VLRA value
cn(i,j),c = 1, ∀n ∈ GV , (i, j) ∈ E nV (32)
from O nV and map it on the auxiliary graph based on PLRA,
c∈C P
where the order of the physical link vector is OLP =
Formula (32) ensures that a VL (i, j) mapping occupies the {pl1 , pl2 , . . . , pli , . . . , pl|E P | }. Secondly, all virtual links are
same index of a core on different fiber links from the same mapped to the physical links by employing a way that the
physical path and only one core can be selected. virtual link with the highest VLRA value is preferentially
7) Bidirectional connection constraint: mapped to the physical link with the highest PLRA value on
n
y(m,u) n
= y(u,m) , ∀n ≤ GV , m ∈ V nV , u ∈ V P (33) the nth VN, which can be described as follows:

Formula (33) ensures that a bidirectional connection is pro- V LRAvvli ∈OLnV → P LRAppli ∈OLP (36)
vided between a virtual node m and its mapped physical node
Fig. 6 (d) shows an example in which a VN is mapped to
u.
a physical network, i.e., SDM-ODCN, with the LRA VN
8) Inter-core crosstalk threshold constraint:
mapping approach. VL mapping may have three cases as
f,c
XTk,l ≥ XTmax , ∀ (k, l) ∈ EP , c ∈ CP , f ∈ FP follows:
X X Case1: Two virtual nodes of a VL (i, j) are mapped
φn,f,c n
(i,j),(k,l) ≤ M × F s(i,j) − 1, (34)
to AG with PLRA at the same time. In this case, if the
c∈C P w∈ f,f +F sn
h i
(i,j) providing spectrum resources of a PL must meet the bandwidth
∀n ∈ GV , (i, j) ∈ E nV , (k, l) ∈ E P (35) requirements of the VL mapping, we can select this selected
link as the mapped PL of this VL on auxiliary graph with
The formula (34) ensures that the inter-core crosstalk XT f,c
k,l PLRA.
of each physical link does not exceed the crosstalk threshold Case2: One of virtual nodes for a VL (i, j) has been
XT max in the SDM-ODCN. The formula (35) is used to already mapped to a PN. This VL is mapped by extended a
reduce the computational complexity. PL from a mapped PN to a non-mapped PN. The selected PL
on the constructed auxiliary graph based on the PLRA value
VII. H EURISTIC VN M APPING A PPROACHES is the highest PLRA value. If available spectrum resources are
In this section, we propose VN mapping approaches with not enough on this physical link, we must search for another
the link resource availability and the node resource availability physical link.
to improve the network profit in SDM-ODCNs. Traditional Case3: Neither of two virtual nodes of a VL (i, j) is mapped
VN mapping approaches only consider the single resources of to the auxiliary graph based on the PLRA value. At this
virtual nodes or virtual links [29]. A VN mapping approach situation, we must search for a PL with the highest PLRA
with a virtual node priority ensures the success ratio of value that has not already been mapped. If the providing spec-
VN mapping, but ignores the network cost of VN mapping. trum resources on the PL satisfy the bandwidth requirements
A VN mapping approach with a virtual link priority reduces of this VL, we can employ this PL as the mapped PL on
the network cost network cost of VN mapping, but doesn’t auxiliary graph based on the PLRA value. The VN with the
improve the success ratio of VN mapping. largest computing resource requirements is mapped to the PN
However, the total network profit of VN mapping is deter- with the largest computing resources provisioning. Otherwise,
mined by the number of successfully mapped VNs and the we should find another PL mapping with a higher PLRA value.
network cost. Therefore, the network cost of VN mapping The LRA VN mapping approach is described as Approach 1.
should be considered in order to maximize the revenue,
as well as the number of successful VN mappings. Link B. NRA VN Mapping Approach
resource availability and node resource availability are defined The NRA VN mapping approach also achieves by using the
by balancing the number of VN mappings and the network way that the virtual nodes with the highest VNRA values are
cost. It considers the usage state of link bandwidth resources, preferentially mapped to the physical node with the highest
node computing resource, and distance between node pairs. PNRA value. We sort the virtual nodes with the VNRA value
Then according to the definition to design the VN mapping in a descending order, andnthis order is marked as the vir-
approaches with both link resource availability and node
o
tual link vector, ON nV = vn1 , vn2 , . . . , vni , . . . , vn|V n | ,
resource availability, we can achieve the purpose of the cost- V

benefit optimization. where |V nV | is the total number of virtual nodes of the nth
VNs. We then take the virtual nodes with the highest VNRA
value from ON nV and attempt to map it to the PL with the
A. LRA VN Mapping Approach
highest PNRA value. The order of the physical node vector
The LRA VN mapping approach achieves VN mapping is ON P = {pn1 , pn2 , . . . , pni , . . . , pn|V P | }, where |V P | is
by using the way that the virtual link with the highest the total number of PNs. We then check whether the spectrum
VLRA value is preferentially mapped to the physical link resources are sufficient on the corresponding PL, and find the
with the highest PLRA value for each VN. We firstly sort the next PN if the link is blocked. Finally, we can map all virtual
virtual links in a descending order with their VLRA values nodes on the nth VN to the physical nodes such that the virtual
and record this order in the virtual link vector, OLnV = nodes with the highest VNRA values are preferentially mapped
{vl1 , vl2 , . . . , vli , . . . , vl|E n | } for the nth VN, where |E nV | to the physical node with the highest PNRA value.
V
Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 CHEN et al.: CROSSTALK-AWARE VN MAPPING IN SDM-ODCNs 3535

Approach 1 LRA VN Mapping Approach Approach 2 NRA VN Mapping Approach

Input: A SDM-ODCN, Gp = {Vp , Ep , Op , Fp , Cp }, a set of VNs, Input: A SDM-ODCN, Gp = {Vp , Ep , Op , Fp , Cp }, a set of
and Gv = {gvn |n = 1, 2, . . . , |Gv |}. VNs, and Gv = {gvn |n = 1, 2, . . . , |Gv |}.
Output: Total network profit after a given of VN mapping.
Output: Total network profit after a set of VN mapping.
1: for all VNs, Gv = {gvn |n = 1, 2, · · · , |GV |} do
2: get the nth VN, gvn = {V n n n n 1: for all VNs Gv = {gvn |n = 1, 2, · · · , |GV |} do
V , E V , C V , B V } ∈ G V , Gv =
{gvn |n = 1, 2, . . . , |Gv |}; 2: Get nth VN gvn = {V nV , E nV , C nV , B nV } ∈ GV , GV =
3: Construct auxiliary graph (AG) based on the PLRA value for the {gvn |n = 1, 2, · · · , |GV |}
current physical network, GP ; 3: Construct auxiliary graph based on PNRA for the current
4: for all descending the order based on the VLRA value in the physical network;
virtual link vector, OLn V = {vl1 , vl2 , . . . , vli , . . . , vl|E n | } on the n
nth VN do
V 4:
n for all descending order of o the virtual nodes, ON V =
5: Get a VL (i, j) from the lowest value to the highest one in vn1 , vn2 , . . . , vni , . . . , vn|V n | on the nth VN based on the
V
virtual link vector OLn V; VNRA value do
6: Case1: two virtual nodes i and j are mapped simultaneously; 5: Get a vni from the lowest order to the highest one in
Assume that a VL (i, j) was mapped to a PL (k, l);
7: if the F sn
ON nV ;
i,j continuous FSs with the XT value below
the threshold XTmax on the mapped PL (k, l) can be allocated 6: for all get a PN pni with the highest PNRA on the
successfully then do constructed AG;
8: Map a VL (i, j) to a PL (k, l) on auxiliary graph, and 7: if cnv i ≤ ckP then do
update the status of the computing resources of PNs and the spectrum 8: Get another virtual node j from V nV −order and
resources of PLs;
9: Calculate network cost for a PL (k, l).
another PN l on the constructed AG according to the highest
10: end if VNRA versus the highest PNRA, VNRA →PNRA;
n
11: Case2: One of two virtual nodes i and j is mapped; 9: if cv j ≤ clP then do
12: A virtual node i is not mapped yet, and a virtual node j 10: Search for the continuous F sni,j FSs with the
was already mapped to a PN l; XT value below the threshold XTmax along the selected PL
13: for all k ∈ V p that is not mapped do
14: Chosen the path with the highest PLRA value between
(k, l) on the constructed AG based on the PNRA value and
the PNs k and l that the virtual node j mapped to; allocate FSs on the PL (k, l) of the VL (i, j);
15: if cn k
v ≤ cP then do
i 11: Map a VL (i, j) to a PL (k, l) on the constructed
16: Search for F sn i,j continuous FSs with the XT below AG based on PNRA, and update the status of the computing
the threshold XTmax along the selected PL (k, l) on AG based on resources of PNs and the spectrum resources of PLs
PLRA and allocate FSs on a PL (k, l) of a VL (i, j);
17: Map a VL (i, j) to a PL (k, l) on AG based on PLRA
12: Calculate the network cost for a PL (k, l)
value, and update the status of the computing resources of PNs and 13: end if
the spectrum resources of PLs; 14: end if
18: Calculate the network cost for a PL (k, l). 15: end for
19: end if 16: end for
20: end for
21: Case3: two virtual nodes i and j are not mapped yet;
17: If all virtual nodes on the nth VN are successfully
22: for all get a PL (k, l) with the highest PLRA value mapped, we can calculate the network profit of a set of VNs.
on AG do Otherwise, the VN mapping is blocked.
23: A virtual node i has greater computing resource require- 18: end for
ments than the virtual node j, and a PN k is the one with the largest
computing resources providing of the two PNs;
nj
24: if cn k
v ≤ cP and cv ≤ cP then do
i l

25: Search for continuous F sn C. Random (RAN) VN Mapping Approach

i,j FSs with the XT below
the threshold XTmax along the selected PL (k, l) on AG based the To better evaluate the ILP model and the proposed heuristic
VLRA value and allocate FSs with the bandwidth requirements on LRA and NRA VN mapping approaches, a random VN
the PL (k, l) of the VL (i, j);
26: Map a VL (i, j) to a PL (k, l) on AG based on the PLRA value, mapping approach is introduced for comparison. A PN with
and update the status of the computing resources of PNs and the sufficient computing resources is randomly found to map for
spectrum resources of PLs; any virtual node in the SDM-ODCN. Dijkstra’s algorithm
27: Calculate the network cost for a PL (k, l). searches for the shortest path between the two selected PNs in
28: end if order to achieve the VN mapping. If the spectrum resources
29: end for
30: end for are insufficient on a PL, the next node is randomly selected.
31: If all VLs on the nth VN are successfully mapped, we can The RAN VN mapping approach is described as follows.
calculate the network profit for a set of VNs. Otherwise, the VN
mapping is blocked. RAN vi∈V nV → RAN pk∈V P (38)
32:end for
where RAN vi∈V nV is a set of virtual nodes on the nth VN and
RAN pk∈V P is a set of physical nodes. We first randomly map
V N RAvvni ∈ON nV → P N RAppni ∈ON P (37) the virtual nodes to the PNs. We search for the continuous
F sni,j frequency slots with the XT value below the threshold
Fig. 8 (d) shows an example of VN mapping based on NRA. XTmax on the reserved PL(k, l) on the auxiliary graph and
The specific steps are shown in Approach 2: allocate FSs on the PL(k, l) of a VL(i, j). We then map a

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 3536 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 72, NO. 6, JUNE 2024

VL(i, j) to a PL(k, l), and update the status of the computing

resources of the physical node and the spectrum resources of a
PL. If all virtual nodes are successfully mapped on the nth VN,
we calculate the network profit of a set of VNs. Otherwise,
the VN mapping is blocked.

VIII. C OMPLEXITY A NALYSIS

The computational complexity of the ILP model depends
on the computational dimension space of variables. So the
ILP model depends on the variables, y(i.s) n
, φn,f,c
(i.j),(k,l) ,
xn(i.j),(s,d) and cn(i.j),c , where their computational com-
plexity is O(|Gv | × |Vvn | × |Vp |), O(|Gv | × |Env | × Fig. 9. The 6-node topology.
|Ep |×|Cp |×|Fp |), O (|Gv | × |Env | × |Ep | × (|Ep | − 1)) and
O (|Gv | × |Env | × |Cp |), respectively. Thus, the total compu-
tational complexity of the ILP model is: VN is same as the LRA VN mapping approach. Thus the total
computational complexity of the NRA VN mapping approach
O(|Gv | × |V nv | × |V p | + |Gv | × |E nv | × |E p | × |C p | is:
× |F p | + |Gv | × |E nv |
O(|V nv | + (|V p | × (|V p | − 1) + |E nv |) × (|E p | + |V p |
× |E p | × (|E p | − 1) + |Gv | × |E nv | × |C p |) (39)
× log |V p |) + |E nv | × |C p | × |F p | − F sni,j × F sni,j

The Dijkstra algorithm is used to establish transmission × log |V p − 1|) (41)

path for the VN mapping approach. The computational com-
plexity computing transmission path is O(|E p | + |V p | log The RAN VN mapping approach is based on the virtual
|V p |), where |E p | and |V p | are the number of physical nodes that are randomly mapped to the physical nodes by
links and physical nodes, respectively. There are |V p | × checking computing resources on both virtual nodes and
(|V p | − 1) node-pairs in the SDM-ODCN. So the compu- physical nodes. So its computational complexity is determined
tational complexity of constructing the auxiliary graph based by the Dijkstra algorithm. However, the RAN doesn’t need to
on the LRA value is O(|V p | × (|V p | − 1) ×(|E p | + determine the mapping order of virtual nodes, thus the total
|V p | log |V p |)). The computational complexity of deter- computational complexity of the RAN VN mapping approach
mining virtual link mapping order in LRA VN mapping is:
approach is O(|E nv |), where |E nv | is the total number of
virtual links on a VN. The Dijkstra algorithm is used to O((|V p | × (|V p | − 1) + |E vn |) × (|E p | + |V p | × log |V p |)
+ |E vn | × |C p | × |F p | − F sni,j × F sni,j × log |V p − 1|)


map virtual links on the auxiliary graph based on the PLRA
value, and its computational complexity is O(|E p | + |V p | (42)
log |V p |). So the computational complexity of mapping all
virtual links on a VN is O (|E nv | × (|E p | + |V p | × log |V p |)).
IX. S IMULATION AND R ESULTS
In additional, we consider that the number of fiber links is
approximate for each path. In this section, we evaluate the network performance of

The computational complexity

is
O |C p | × |F p | − F sni,j × F sni,j × log |V p − 1| along one the ILP model and heuristic LRA and NRA VN mapping
transmission path for each virtual link when allocating spec- approaches. In addition, the RAN mapping approach is used
trum resources, where |C p |, |F p |, and F sni,j are the total cores for the comparison approach. In the simulation process,
of each fiber link, the number of frequency slots on each fiber a single line rate with 40Gbps (25GHz) is adopted to carry
core, and the bandwidth requirements of a virtual link (i, j) the bandwidth requirements. Several metrics proposed are
on the nth VN. So the total computational complexity of the obtained in simulation results.
LRA VN mapping approach is:
O(|E nv | + (|V p | × (|V p | − 1) + |E nv |) × (|E p | + |V p | A. Simulation Results in a 6-Node Network
× log |V p |) + |E nv | × |C p | × |F p | − F sni,j × F sni,j


1) Simulation Conditions: The ILP model and the heuristic
× log |V p − 1|) (40) VN mapping approaches are simulated in the 6-node topology
with 8 bidirectional links with 3 fiber cores and 6 nodes in
The NRA VN mapping approach is based on the PNRA Fig. 9. Each fiber link has three cores, and each fiber core
value to map the physical nodes. The computational complex- can provide 18 frequency slots. The numbers on the circles
ity of determining virtual node mapping order in NRA VN denote the computing resources on each data center that is
mapping approach is O (|V nv |), where |V nv | is the total number near the optical switching node. The numbers on the fiber link
of virtual nodes on a VN. For the NRA VN mapping approach, represent the distance between two optical switching nodes
we also use the Dijkstra algorithm to establish a transmission with km. The required computing resources are within [3, 5]
path and map virtual links on AG based on PNRA. The units for each virtual node. The bandwidth requirements are
computational complexity of mapping all virtual links on a evenly distributed within [35, 45] Gbps for each virtual link.

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 CHEN et al.: CROSSTALK-AWARE VN MAPPING IN SDM-ODCNs 3537

Fig. 11. Average network profit under different number of virtual nodes.
Fig. 10. Average network profit under different number of virtual networks.

We set the value of the equilibrium factor as θ = 1.

Considering the network performance under different number
of VNs, we give a set of VNs and each VN consists of
3 nodes and 3 links in advance. A data center near the
physical node can provide computing resources within [30,
60] units. In order to verify the network performance under
the different number virtual nodes, a VN consists of 2 virtual
nodes with 1 virtual link, 3 virtual nodes with 3 virtual links,
4 virtual nodes with 4 virtual links, and 5 virtual nodes with
5 virtual links, respectively. Moreover, we enlarge the number
of computing resources provisioning from 60 to 80 units on
each data center under different number of virtual nodes. For Fig. 12. Spectrum occupancy ratio under different number of virtual
networks.
each data point, the ILP model, the LRA, NRA, and RAN VN
mapping approaches are applied to the 10 sets of VNs, and
the results of the mapping are averaged. mapping approach obtains the higher average network profit
2) Simulation Results: The average network profit of the compared with the NRA and RAN VN mapping approaches.
ILP model, the LRA, NRA, and RAN VN mapping approaches We can obtain the spectrum occupancy ratio of the ILP
under different VN numbers is shown in Fig. 10. We can see model, the LRA, NRA, and RAN VN mapping approaches
that the ILP model achieves the highest average network profit under different number of VNs in Fig. 12. We can see that
than the LRA, NRA, and RAN VN mapping approaches. The the ILP model has lower spectrum occupancy ratio compared
reason is that the ILP model employs global optimization for to the LRA, NRA, and RAN VN mapping approaches. Obvi-
a given set of VNs in advance. We also see that the LRA VN ously, the ILP model can achieve the optimized objective.
mapping approach obtains higher average network profit than Moreover, the LRA VN mapping is better than the NRA and
the NRA and RAN VN mapping approaches. The reason is RAN VN mapping approaches and is much closer to optimal
that the LRA VN mapping approach maps the VLs with the solutions of the ILP model. Therefore, the spectrum occupancy
highest VLRA value to the PLs with the highest PLRA value. ratio is greatly reduced by employing the ILP model and the
Moreover, the average network profit of the LRA VN mapping LRA VN mapping approach. In Fig. 13, the simulation results
approach is much closer to the ILP model and significantly show that the he ILP model achieves the lowest spectrum
better than the NRA and RAN VN mapping approach. The occupancy ratio compared with the LRA, NRA, and RAN
LRA mapping approach has lower 15.8% average network VN mapping approaches. Moreover, the LRA VN mapping
profit compared with the ILP model. The LRA or NRA VN approach has the lower spectrum occupancy ratio than the
mapping approach has the same average network profit. This is NRA, and RAN VN mapping approaches.
because that there are no VNs blocking when mapping a given
set of VNs. The transmission path of the LRA or NRA VN
mapping approach may be the same, where the number of the B. Simulation Results in a 14-Node Network
optical regenerators has the same configuration for the LRA or 1) Simulation Conditions: We employ NSFNET in Fig. 14
NRA VN mapping approach. In Fig. 11, the simulation results to verify the performance of the LRA, NRA, and RAN VN
show that the ILP model has the highest average network profit mapping approaches. We consider seven cores on each fiber
than the LRA, NRA, and RAN VN mapping approaches under link, and each fiber core has 130 frequency slots. The band-
different number of virtual nodes. Furthermore, the LRA VN width requirements are required to be evenly distributed with

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 3538 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 72, NO. 6, JUNE 2024

Fig. 13. Spectrum occupancy ratio under different number of virtual nodes.
Fig. 15. Average network profit under different number of virtual networks.

Fig. 14. The NSFNET topology.

Fig. 16. Average network profit under different number of virtual nodes.
the range [200, 340] Gbps for a virtual link. The computing
resources are within a range of [5, 8] units for a virtual
node. Each data centers provide computing resources with employs the way that the VLs with the highest VLRA value
a range [150, 250] units. A set of VNs is generated, where is mapped to PLs with the PLRA value. Moreover, the NRA
each VN has 5 virtual nodes and 7 virtual links when we VN mapping approach can achieves higher network profit by
consider the network performance under different number of mapping the virtual nodes with the highest VNRA value to the
VNs. Moreover, we also consider the effect of the different physical nodes with the highest PNRA value compared to the
number of virtual nodes on the network performance. Here, RAN VN mapping approach. In Fig. 16, we can also see that
we consider that there are 1, 3, 4, 5, 7, and 9 fiber links of both the LRA and NRA VN mapping approaches have better
a VN under 2, 3, 4, 5, 6, and 7 virtual nodes, respectively. average network profit compared with the RAN VN mapping
For each data point, we applied 100 sets of VNs to do the approach under different number of virtual nodes.
simulation and the simulation results are obtained based on In Fig. 17, compared to the NRA and RAN VN mapping
the average of the 100 sets. approaches, the LRA VN mapping approach achieves the
2) Simulation Results: Generally, for the ILP model, it is lowest average crosstalk. The reason is that the LRA VN
unsuited for mapping a large number of VNs to large-scale mapping approach maps the VLs with the highest VLRA
SDM-ODCN. Therefore, we only employ the heuristic VN value to PLs with the PLRA value. We also observe that
mapping approaches to verify the performance of the LRA, the NRA VN mapping approach is lower average crosstalk
NRA, and RAN VN mapping approaches in the simula- than the RAN VN mapping approach. The reason is that the
tion. The metrics include the average network profit, average NRA VN mapping approach maps the virtual nodes with
crosstalk, rejection ratio, spectrum occupancy ratio, and fre- the highest VNRA value to PNs with the highest PNRA
quency fragment, which are analyzed are follows. value. The average crosstalk increases for three VN mapping
The LRA, NRA, and RAN VN mapping approaches obtain approaches with the number of VNs increasing. Fig. 18 shows
the average network profit under different number of VN the average crosstalk of different VN mapping approaches
mapping in the NSFNET topology in Fig. 15. We can see that under different number of virtual nodes. We can see that
the LRA and NRA VN mapping approaches improve 14.0% the LRA VN mapping approach achieves the lower average
and 9.1% average network profit compared with the RAN crosstalk compared with the NRA and RAN VN mapping
VN mapping approach, since the LRA VN mapping approach approaches.

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 CHEN et al.: CROSSTALK-AWARE VN MAPPING IN SDM-ODCNs 3539

Fig. 17. Average crosstalk under different number of virtual networks. Fig. 20. Rejection ratio of VNs under different number of virtual nodes.

Fig. 18. Average crosstalk under different number of virtual nodes.

Fig. 21. Spectrum occupancy ratio under different number of virtual
networks.

we can verify that the RAN and NRA VN mapping approaches

have higher rejection ratio of VNs than the LRA VN mapping
approach under different number of virtual nodes. The is
because the LRA VN mapping approach preferentially maps
virtual links with higher LRA value, and the average hops of
the transmission paths are less than those of the NRA VN
mapping approaches.
In Fig. 21, the LRA VN mapping approach has lower
spectrum occupancy ratio compared to the NRA and RAN
VN mapping approaches. Obviously, we find that the LRA
VN mapping approach not only is lower rejection ratio, but
also has lower spectrum occupancy ratio compared to the NRA
Fig. 19. Rejection ratio of VNs under different number of virtual networks. and RAN VN mapping approaches. The reason is that the hop
counts of mapping paths for the LRA VN mapping approach
are less than those of the NRA and RAN VN mapping
The rejection ratio of VNs is shown under different number approaches. Furthermore, in Fig. 22, we can observe that
of VNs in Fig. 19. Compared with the RAN VN mapping the LRA VN mapping approach achieves the lower spectrum
approach, both LRA and NRA VN mapping approaches have occupancy ratio compared to the NRA and RAN VN mapping
lower rejection ratio of VNs, since we map the VNs to the approaches under different number of virtual nodes.
physical network by using both the VLs with the highest In Fig. 23, we see that both the LRA and NRA VN mapping
VLRA value and the virtual nodes with the highest VNRA approaches achieve a much lower spectrum fragmentation ratio
value. Moreover, the LRA and NRA VN mapping approaches compared to the RAN VN mapping approach. This is because
have no rejection ratio of VNs, since they are enough spectrum the spectrum resources can be allocated more regularly by
resources and computing resources provisioning. In Fig. 20, using both LRA and NRA VN mapping for the physical links,

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 3540 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 72, NO. 6, JUNE 2024

Fig. 22. Spectrum occupancy ratio under different number of virtual nodes.

Fig. 24. Spectrum fragmentation ratio under different number of virtual

nodes.

X. C ONCLUSION
In this paper, in order to improve the network profit and
spectrum efficiency, we investigated the crosstalk-aware VN
mapping problems in SDM-ODCNs. We proposed two novel
mapping parameters about both link resource availability and
node resource availability in order to realize effective VN
mapping. Furthermore, we proposed an ILP model, the LRA
and NRA VN mapping approaches to improve the network
profit in SDM-ODCNs. On one hand, we can find that the
simulation results of our proposed LRA VN mapping approach
are greatly close to the solutions of the optimal ILP model.
Fig. 23. Spectrum fragmentation ratio under different number of virtual On the other hand, the simulation results show that the
networks.
LRA and NRA VN mapping approaches improve 14.0% and
9.1% average network profit compared with the RAN VN
mapping approach in a 14-node SDM-ODCN. Meanwhile, the
which makes full use of the spectrum resources. Obviously, LRA VN mapping approach not only improves the network
we can reduce the spectrum fragmentation ratio by using the profit, but also suppresses the average crosstalk and reduces
LRA and NRA VN mapping approaches when assigning spec- both the rejection ratio of VNs and the spectrum fragmen-
trum resources. Moreover, the LRA VN mapping approach has tation ratio compared to the NRA and RAN VN mapping
lower spectrum fragmentation ratio than the NRA VN mapping approaches.
approach. This is because the LRA VN mapping approach
chooses the transmission path with fewer hops compared to
R EFERENCES
the NRA and RAN mapping approaches.
Fig. 24 shows the spectrum fragmentation ratio of different [1] N. M. K. Chowdhury and R. Boutaba, “A survey of network virtualiza-
tion,” Comput. Netw., vol. 54, no. 5, pp. 862–876, 2010.
VN mapping approaches under different number of virtual [2] L. Gong and Z. Zhu, “Virtual optical network embedding (VONE) over
nodes. Compared with the NRA and RAN VN mapping elastic optical networks,” J. Lightw. Technol., vol. 32, no. 3, pp. 450–460,
approaches, the LRA VN mapping approach always has lower Feb. 1, 2014.
[3] Q. Zhang, X. Zhang, X. Gong, and L. Guo, “Crosstalk-avoid virtual opti-
spectrum fragmentation ratio. The reason is that the LRA cal network embedding over elastic optical networks with heterogeneous
VN mapping approach uses the way that the VLs with the multi-core fibers,” J. Lightw. Technol., vol. 40, no. 24, pp. 7687–7700,
highest VLRA value is mapped to PLs with the PLRA value. Dec. 15, 2022.
[4] L. Zhang, J. Chen, E. Agrell, R. Lin, and L. Wosinska, “Enabling tech-
Moreover, the NRA VN mapping approach always has lower nologies for optical data center networks: Spatial division multiplexing,”
spectrum fragmentation ratio than the RNA VN mapping J. Lightw. Technol., vol. 38, no. 1, pp. 18–30, Jan. 1, 2020.
approach before the number of virtual nodes is 5. However, [5] M. Pham, D. B. Hoang, and Z. Chaczko, “Congestion-aware and energy-
the NRA VN mapping approach always has higher spectrum aware virtual network embedding,” IEEE/ACM Trans. Netw., vol. 28,
no. 1, pp. 210–223, Feb. 2020.
fragmentation ratio than the RNA VN mapping approach after [6] M. Berman et al., “GENI: A federated testbed for innova-
the number of virtual nodes is 5. The reason is that the tive network experiments,” Comput. Netw., vol. 61, pp. 5–23,
RNA VN mapping approach has more rejection ratio of VNs Mar. 2014.
[7] R. Sherwood et al., “Carving research slices out of your production
than the NRA VN mapping approach with the virtual nodes networks with OpenFlow,” ACM SIGCOMM Comput. Commun. Rev.,
increasing. vol. 40, no. 1, pp. 129–130, Jan. 2010.

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 CHEN et al.: CROSSTALK-AWARE VN MAPPING IN SDM-ODCNs 3541

[8] M. Jinno et al., “Distance-adaptive spectrum resource allocation in Bowen Chen (Member, IEEE) received the Ph.D.
spectrum-sliced elastic optical path network [topics in optical communi- degree from the Beijing University of Posts and
cations],” IEEE Commun. Mag., vol. 48, no. 8, pp. 138–145, Aug. 2010. Telecommunications, Beijing, China, in 2014. He is
[9] S. Fujii, Y. Hirota, H. Tode, and K. Murakami, “On-demand spectrum currently an Associate Professor with the School of
and core allocation for reducing crosstalk in multicore fibers in elastic Electronic and Information Engineering, Soochow
optical networks,” IEEE/OSA J. Opt. Commun. Netw., vol. 6, vol. 12, University. He is the author/coauthor of over 70 ref-
pp. 1059–1071, Dec. 2014. ereed technical articles. His current research interests
[10] N. Benzaoui, S. Sahoo, and S. Bigo, “Deterministic latency networks: include optical network design, network survivabil-
The enabler of edge data center synchronous operation [Invited],” J. Opt. ity, optical network virtualization, and edge-cloud
Commun. Netw., vol. 13, no. 9, pp. D115–D125, Sep. 2021. computing networks.
[11] E. Ghazisaeedi and C. Huang, “GreenMap: Green mapping of
MapReduce-based virtual networks onto a data center network and
managing incast queueing delay,” Comput. Netw., vol. 112, pp. 345–359,
Jan. 2017.
[12] M. P. Gilesh, S. D. Madhu Kumar, and L. Jacob, “Resource availability–
aware adaptive provisioning of virtual data center networks,” Int. J. Netw.
Manage., vol. 29, no. 2, Mar. 2019. Wenwen Zheng received the B.E. degree in com-
[13] H. Jiang, Y. Wang, L. Gong, and Z. Zhu, “Availability-aware survivable munication engineering from Ludong University,
virtual network embedding in optical datacenter networks,” J. Opt. Yantai, China, in 2020. She is currently pursuing
Commun. Netw., vol. 7, no. 12, pp. 1160–1171, Dec. 2015. the master’s degree in information and communica-
[14] X. Li, H. Wang, B. Ding, X. Li, and D. Feng, “Resource allocation tion engineering with Soochow University, Suzhou,
with multi-factor node ranking in data center networks,” Future Gener. China. Her current research interests include virtual
Comput. Syst., vol. 32, pp. 1–12, Mar. 2014. optical network mapping and network resource opti-
[15] A. Jahani, L. M. Khanli, M. T. Hagh, and M. A. Badamchizadeh, “Green mization.
virtual network embedding with supervised self-organizing map,” Neu-
rocomputing, vol. 351, pp. 60–76, Jul. 2019.
[16] W. Zhen et al., “Virtual optical network mapping approaches with inter-
core crosstalk in space division multiplexing elastic optical data center
networks,” in Proc. 19th Int. Conf. Opt. Commun. Netw. (ICOCN),
Aug. 2021, pp. 1–4.
[17] J. M. D. Mendinueta, S. Shinada, Y. Hirota, H. Furukawa, and N. Wada,
“High-capacity super-channel-enabled multi-core fiber optical switching Jun Yu received the B.E. degree from the School
system for converged inter/intra data center and edge optical networks,” of Electronic and Information Engineering, Soochow
IEEE J. Sel. Topics Quantum Electron., vol. 26, no. 4, pp. 1–13, University, Suzhou, China, in 2020, where he is cur-
Jul. 2020. rently pursuing the master’s degree in electronic and
[18] Y. Zhao, L. Hu, R. Zhu, X. Yu, X. Wang, and J. Zhang, “Crosstalk-aware communication engineering. His current research
spectrum defragmentation based on spectrum compactness in space interests include service function chain mapping
division multiplexing enabled elastic optical networks with multicore topology design and mapping algorithm research.
fiber,” IEEE Access, vol. 6, pp. 15346–15355, 2018.
[19] S. Ba, B. C. Chatterjee, and E. Oki, “Defragmentation scheme based on
exchanging primary and backup paths in 1+1 path protected elastic opti-
cal networks,” IEEE/ACM Trans. Netw., vol. 25, no. 3, pp. 1717–1731,
Jun. 2017.
[20] M. R. Rahman and R. Boutaba, “SVNE: Survivable virtual network
embedding algorithms for network virtualization,” IEEE Trans. Netw.
Service Manage., vol. 10, no. 2, pp. 105–118, Jun. 2013.
[21] B. Chen, J. Zhang, W. Xie, J. P. Jue, Y. Zhao, and G. Shen,
Danyang Zheng (Member, IEEE) received the B.S.
“Cost-effective survivable virtual optical network mapping in flexi-
degree in computer science from the University
ble bandwidth optical networks,” J. Lightw. Technol., vol. 34, no. 10,
of Electronic Science and Technology of China,
pp. 2398–2412, May 15, 2016.
Chengdu, China, in 2016, and the Ph.D. degree
[22] A. Jarray and A. Karmouch, “Cost-efficient mapping for fault-tolerant in computer science from Georgia State Univer-
virtual networks,” IEEE Trans. Comput., vol. 64, no. 3, pp. 668–681, sity, Atlanta, GA, USA, in 2021. He is currently
Mar. 2015. an Assistant Professor with Soochow University,
[23] H. Yu, T. Wen, H. Di, V. Anand, and L. Li, “Cost efficient virtual Suzhou, China. His research interests include net-
network mapping across multiple domains with joint intra-domain and work function virtualization, network reliability and
inter- domain mapping,” Opt. Switching Netw., vol. 14, pp. 233–240, security, in-network computation, and combinational
Aug. 2014. optimization. He has published more than 40 tech-
[24] T. Hayashi, T. Taru, O. Shimakawa, T. Sasaki, and E. Sasaoka, “Design nical works in his research fields. He is serving as an Associate Editor for
and fabrication of ultra-low crosstalk and low-loss multi-core fiber,” Opt. IEEE B IG DATA M INING AND A NALYTICS.
Exp., vol. 19, no. 17, pp. 16576–16592, 2011.
[25] Q. Chen et al., “Virtual optical network mapping approaches in space-
division-multiplexing elastic optical data center networks,” J. Lightw.
Technol., vol. 40, no. 12, pp. 3515–3529, Jun. 15, 2022.
[26] Q. Yao et al., “Core and spectrum allocation based on association rules
mining in spectrally and spatially elastic optical networks,” IEEE Trans.
Hong Chen received the master’s degree in electron-
Commun., vol. 69, no. 8, pp. 5299–5311, Aug. 2021.
ics and communication engineering from Soochow
[27] M. Koshiba, K. Saitoh, K. Takenaga, and S. Matsuo, “Multi-core fiber
University, Suzhou, China, in 2008. She is currently
design and analysis: Coupled-mode theory and coupled-power theory,”
an Associate Professor with the School of Elec-
Opt. Exp., vol. 19, no. 26, pp. B102–B111, 2011.
tronics and Information, Soochow University. Her
[28] K. Takenaga et al., “An investigation on crosstalk in multi-core fibers current research interests include optical networking
by introducing random fluctuation along longitudinal direction,” IEICE technology and network resource optimization.
Trans. Commun., vol. E94-B, no. 2, pp. 409–416, 2011.
[29] B. Chen, Y. Zhao, and J. Zhang, “Energy-efficient virtual optical network
mapping approaches over converged flexible bandwidth optical networks
and data centers,” Opt. Exp., vol. 23, no. 19, pp. 24860–24872, 2015.

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.
 3542 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 72, NO. 6, JUNE 2024

Mingyi Gao received the Ph.D. degree in informa- Jason P. Jue (Senior Member, IEEE) received the
tion and communication engineering from Shanghai B.S. degree in electrical engineering and computer
Jiao Tong University, Shanghai, China, in 2007. science from the University of California at Berkeley
She was a Research Fellow with the Japan Society in 1990, the M.S. degree in electrical engineering
for the Promotion of Science (JSPS), The Univer- from the University of California at Los Angeles in
sity of Tokyo, Tokyo, Japan, from 2007 to 2009. 1991, and the Ph.D. degree in computer engineering
From 2009 to 2013, she was with the National Insti- from the University of California at Davis in 1999.
tute of Advanced Industrial Science and Technology, He is currently a Professor with the Department
Tsukuba, Japan. She is currently an Associate of Computer Science, The University of Texas at
Professor with the School of Electronic and Informa- Dallas. His current research interests include optical
tion Engineering, Soochow University. Her current networks and network survivability.
research interests include coherent optical communication, multicarrier PON,
digital signal processing, machine learning, and FPGA.

Gangxiang Shen (Senior Member, IEEE) received

the B.Eng. degree from Zhejiang University, China,
the M.Sc. degree from Nanyang Technological Uni-
versity, Singapore, and the Ph.D. degree from the
Weiguo Ju received the Ph.D. degree from the University of Alberta, Canada, in 2006. He was a
Beijing University of Posts and Telecommunica- Lead Engineer with Ciena, Linthicum, MD, USA.
tions, Beijing, China, in 2013. He is currently a He was an Australian ARC Post-Doctoral Fellow
Senior Engineer with the Institute of ICT Technol- with the University of Melbourne. He is cur-
ogy, China Information Consulting and Designing rently a Distinguished Professor with the School of
Institute Company Ltd. His current research inter- Electronic and Information Engineering, Soochow
ests include software-defined optical networks and University, China. He has authored and coauthored
edge-cloud computing networks. more than 200 peer-reviewed technical articles. His research interests include
integrated optical and wireless networks, spectrum-efficient optical networks,
and green optical networks. He is a fellow of OSA. He is a Voting Member of
the IEEE ComSoc Strategic Planning Standing Committee from 2018 to 2019.
He is currently a TPC Member of OFC and ECOC. He was a recipient of
the Young Researcher New Star Scientist Award in the 2010 Scopus Young
Researcher Award Scheme in China. He was a recipient of the Izaak Walton
Killam Memorial Award from the University of Alberta and the Canadian
NSERC Industrial Research and Development Fellowship. He was selected
Pin-Han Ho (Fellow, IEEE) received the Ph.D. as a Highly Cited Chinese Researcher by Elsevier from 2014 to 2018. He was
degree from Queens University, Kingston, ON, elected as an IEEE ComSoc Distinguished Lecturer from 2018 to 2019.
Canada, in 2002. He is currently a Full Professor He was the Secretary of the IEEE Fiber-Wireless (FiWi) Integration Sub-
with the Department of Electrical and Computer Technical Committee. He is a Lead Guest Editor for the Special Issue on
Engineering, University of Waterloo. He is the Next-Generation Spectrum-Efficient and Elastic Optical Transport Networks
author/coauthor of over 400 refereed technical arti- of IEEE J OURNAL ON S ELECTED A REAS IN C OMMUNICATIONS and the
cles, several book chapters, and the coauthor of two Guest Editor for the Special Issue on Energy-Efficiency in Optical Networks
books on internet and optical network survivability. of IEEE J OURNAL ON S ELECTED A REAS IN C OMMUNICATIONS. He was an
His current research interests include broadband Associate Editor of IEEE/OSA J OURNAL OF O PTICAL C OMMUNICATION
wired and wireless communication networks, includ- AND N ETWORKING . He is currently an Associate Editor of IEEE/OSA
ing wireless transmission techniques, mobile system J OURNAL OF L IGHTWAVE T ECHNOLOGY and IEEE N ETWORKING L ET-
design and optimization, and network dimensioning and resource allocation. TERS and an Editorial Board Member of Optical Switching and Networking
He is a Professional Engineer Ontario (PEO). and Photonic Network Communications.

Authorized licensed use limited to: Nanjing Univ of Post & Telecommunications. Downloaded on December 14,2024 at 04:22:25 UTC from IEEE Xplore. Restrictions apply.