Journal of Network and Computer Applications 224 (2024) 103829

Contents lists available at ScienceDirect

Journal of Network and Computer Applications

journal homepage: www.elsevier.com/locate/jnca

Research paper

A service function chain mapping scheme based on functional aggregation in

space-air-ground integrated networks
Peiying Zhang a,b ,∗, Kunkun Yan a , Neeraj Kumar c , Lizhuang Tan d ,∗∗, Mohsen Guizani e ,
Konstantin Igorevich Kostromitin f , Jian Wang g , Jianyong Zhang h,i ,∗∗∗
a
Qingdao Institute of Software, College of Computer Science and Technology, China University of Petroleum (East China), Qingdao, 266580, China
b State Key Laboratory of Integrated Services Networks, Xidian University, Xi’an, 710071, China
c Department of Computer Science and Engineering, Thapar Insitute of Engineering and Technology (Deemed to be University), Patiala, 147004, India
d Key Laboratory of Computing Power Network and Information Security, Ministry of Education, Shandong Computer Science Center (National Supercomputer

Center in Jinan), Qilu University of Technology (Shandong Academy of Sciences), Jinan, 250013, China
e
Machine Learning Department, Mohamed Bin Zayed University of Artificial Intelligence (MBZUAI), Abu Dhabi, 999041, United Arab Emirates
f
Department of Physics of Nanoscale Systems, South Ural State University, Chelyabinsk, 454080, the Russian Federation
g
College of Science, China University of Petroleum (East China), Qingdao, 266580, China
h
Institute of Lightwave Technology, Beijing Jiaotong University, Beijing, 100044, China
i Key Lab of All Optical Network and Advanced Telecommunication of EMC, Beijing Jiaotong University, Beijing, 100044, China

ARTICLE INFO ABSTRACT

Keywords: Being a novel network architecture, the Space-air-ground integrated network (SAGIN) offers advantages such
Space-air-ground integrated network as extensive network coverage and seamless ubiquitous access. However, it also encounters the challenge of
Network function virtualization balancing a growing user demand with limited network service resources. In order to further improve the
Service function chain
utilization rate of network resources, a group of virtual network functions can be connected according to
Aggregation
certain business logic to form a dynamic reconfigurable service function chain. This can provide diversified
and high-quality network services for users. So by aggregating the same type of network function, we can
reduce the cost of service chain mapping. Based on the above ideas, this paper proposes an Aggregation of
Service Function Chain mapping algorithm (A-SFC). The aggregated service function chain topology is mapped
according to an improved isomorphic graph search mapping algorithm, resulting in a deployment solution that
minimizes the mapping cost by aggregating static SFC requests by end-system division. Experimental results
show that the algorithm is able to find an effective mapping scheme under resource constraints. In addition to
reducing computation, the algorithm effectively minimizes the consumption of computational resources while
delivering excellent performance in terms of service delay and mapping cost.

1. Introduction networks and data centers that can make a wide area coverage, real-
time access, security control, on-demand access, real-time information
The rapid development of terrestrial Internet technology has led capabilities (Sheng et al., 2021; Kua et al., 2021). In addition, it
to an increasing demand for expanding the Internet into space and will provide information services for reconnaissance, mapping, re-
establishing a space-air-ground integrated network (SAGIN), driven sources, disaster mitigation, meteorological and marine applications
by the complexity of space missions (Gu et al., 2020). The SAGIN satellites (Alzahrani et al., 2020). At the same time, the rapid devel-
takes communication satellites, relay satellites, navigation satellites and opment of the future Internet of Things, cloud computing network and
other satellites as the main body, and integrate them with terrestrial other new network scenarios, the number of intelligent devices and user
information infrastructure such as fixed and wireless communication demand has also increased. In addition how to make full use of the

∗ Corresponding author at: Qingdao Institute of Software, College of Computer Science and Technology, China University of Petroleum (East China), Qingdao,
266580, China.
∗∗ Corresponding author at: Key Laboratory of Computing Power Network and Information Security, Ministry of Education, Shandong Computer Science Center
(National Supercomputer Center in Jinan), Qilu University of Technology (Shandong Academy of Sciences), Jinan, 250013, China.
∗∗∗ Corresponding author at: Institute of Lightwave Technology, Beijing Jiaotong University, Beijing, 100044, China.
E-mail addresses: [email protected] (P. Zhang), [email protected] (K. Yan), [email protected] (N. Kumar), [email protected]
(L. Tan), [email protected] (M. Guizani), [email protected] (K.I. Kostromitin), [email protected] (J. Wang), [email protected] (J. Zhang).

https://doi.org/10.1016/j.jnca.2024.103829
Received 24 September 2023; Received in revised form 14 December 2023; Accepted 15 January 2024
Available online 30 January 2024
1084-8045/© 2024 Elsevier Ltd. All rights reserved.
P. Zhang et al. Journal of Network and Computer Applications 224 (2024) 103829

Fig. 1. Schematic of SFC dynamic orchestration in SAGIN scenario.

limited computing, storage and bandwidth and other network resources the traffic passes through the VNF carried by the air-space network
and service resources, and in the massive terminal data to achieve node, at this time, the space-air network node has two main roles,
intelligent control, in order to obtain a more efficient and high-quality one is to balance the traffic load, when certain routing nodes of the
network performance and quality of service, has become an urgent terrestrial network are congested with traffic, it can be transmitted
problem to be solved (Shafique et al., 2020). using the space-air node to reduce the traffic load; the other is to
The emergence of software-defined networking (SDN) and network shorten the logical transmission distance and reduce the number of end-
function virtualization (NFV) technologies has greatly alleviated the to-end transmission hops, which is more suitable for communication
aforementioned problem, and SDN and NVF can flexibly deploy net- access and network service requirements in remote and sparse areas.
work functions in virtual machines of cloud services to achieve low-cost Although the SFC solution effectively reduces the operating costs of
and flexible network management. In the aforementioned management network service providers, there are still many problems to be solved in
model, the network function running on a virtual machine is referred the deployment and orchestration of VNFs (Han et al., 2015). The main
to as a virtual network function (VNF). The Service Function Chain issue is how to choose the deployment location and sharing scheme of
(SFC) is a sequential arrangement of VNF instances, where multiple VNF to further optimize the cost and delay under the premise of guaran-
VNFs are interconnected, and user traffic is directed through these teeing the quality of service (QoS) of the user’s network (e.g., to meet
VNFs based on a predefined policy (Van Bemten et al., 2018). As shown the delay and bandwidth requirements) (Yang et al., 2020). Most of
in Fig. 1, this paper proposes a dynamic orchestration architecture the current research works usually consider only the VNF instantiation
for service function chaining in a SAGIN scenario. Among them, the cost or only the bandwidth cost, and adopt the approach that a physical
left part of Fig. 1 shows the process of SAGIN virtualization, which node provides resource allocation for only one virtual node. The above
describes the process of abstracting the underlying physical network SFC deployment scheme is not conducive to flexible resource deploy-
resources composed of air and space integration into virtual network ment in real situations and cannot fully utilize the instantiated function
resources through network virtualization technology. At the physical types, thus severely limiting the application of network virtualization
structure level, the underlying physical network can be divided into technology in SAGIN, while this rigid VNF deployment scheme also
three levels: air, space and ground, which are the space network layer leads to increased mapping costs and wasted resource energy consump-
composed of Low Earth Orbit (LEO) satellites, the near-Earth platform tion (Sun et al., 2020). For this reason, this paper investigates a SFC
layer composed of airplanes and the ground network layer composed mapping scheme based on functional aggregation. This scheme aims to
of local area networks (LANs), cellular mobile networks, etc. Due to the fully utilize the flexible allocation characteristics of virtual resources,
relatively wide coverage of satellites, it can be considered that satellite exploit the similarities between SFC requests originating from the same
nodes have full coverage of the current ground network, while aircraft source node, and leverage VNF instantiation and automatic resource
nodes have partial coverage, and wireless communication is used to expansion. By establishing connection relationships between service
realize connectivity among the above three layers of the network. function chains, it becomes possible to aggregate virtual links with
The right part of Fig. 1 shows the abstracted network topology and the same function type VNF nodes within a set of SFCs that have the
the service function chaining process. The upper layer has the SFC same end nodes. This aggregation forms a single VNF entity with a
information generated by the request, and the lower layer has the real- higher resource demand, ultimately increasing the flexibility of SFC
time state information of the current network topology, and the upper deployment, reducing the cost of SFC mapping, and minimizing service
and lower layers interact with each other to realize the flexible and delays. The purpose is to increase the flexibility of SFC deployment,
dynamic management of VNF mapping, traffic routing, and virtualized reduce the cost of SFC mapping, and reduce the service delay. This
resources in the network topology. VNFs are deployed on some of the article makes the following major contributions:
air, space, and ground network nodes to build a service function chain.
The service function chain resources can be dynamically organized and • Aiming at the problems of traffic congestion and long transmis-
adjusted according to the service requests from different users. When sion distance in SAGIN scenarios, an SFC deployment scheme

2
P. Zhang et al. Journal of Network and Computer Applications 224 (2024) 103829

based on functional aggregation of VNF instances is proposed et al., 2022a) utilized a delay prediction-based SFC mapping algorithm
to minimize the mapping cost while meeting users’ QoS require- to address the delay issue in SAGIN, resulting in enhanced resource
ments. utilization and reduced end-to-end delay. Nonetheless, their approach
• Considering the different service and business resource require- still relies on a one-to-one mapping between virtual nodes and physical
ments, as well as the mobility characteristics of air, sky, and earth nodes.
nodes, we model the SFC mapping problem as a mixed-integer In the study of optimizing the cost of SFC deployment and reducing
linear programming problem (MILP), and establish a solution end-to-end delay, Eramo et al. (2017) conducted a study with the
strategy with the objective of minimizing the mapping cost and objective of reducing energy consumption and minimizing the rejection
delay. of SFC (Service Function Chaining) bandwidth. Chen et al. (2023)
• An SFC mapping algorithm based on functional aggregation is proposed a queue-aware method to dynamically allocate resources and
proposed on the basis of the established MILP model. The com- schedule traffic in the cloud environment, and formulated a latency
parison results with the solutions in the existing literature show limit policy to efficiently meet the service requirements of SFC. Mauro
et al. (2022) introduced two algorithms, OptCNT and OptSearchChain,
that the algorithm in this paper has significant advantages in the
for performance and usability analysis, respectively, aiming to iden-
comprehensive consideration of mapping cost and delay.
tify the optimal latency scheme and fulfill availability requirements
The remainder of this paper is organized as follows. In Section 2, we during SFC deployment in a softwareized IMS environment. Nguyen
provide a review of the related work. In Section 3, we discuss the prob- et al. (2020) developed an SFC orchestration method that considers
lem and introduce the system model. The problem is formally stated demand uncertainty and minimizes frequent resource reconfiguration
and mathematically formulated in Section 4. Our proposed method is by accurately modeling load-independent propagation delays and load-
described in detail in Section 5. The evaluation of the A-SFC algorithm related queuing and processing delays. Although the aforementioned
articles have shown improvements in latency and service request ac-
is presented in Section 6. Finally, we conclude our article in Section 7.
ceptance rate, they do not address resource allocation optimization.
In reality, both VNF selection and routing path, as well as resource
2. Related work
allocation, significantly impact latency performance. However, due to
the heterogeneous nature of SAGIN, a multidimensional network with
Most of the existing studies on SAGIN focus on the architecture
network load balancing and optimized transmission paths, the resource
design, routing path planning and allocation of network resources, such
allocation algorithms proposed in the previous work did not compre-
as communication and computation, with relatively few studies on hensively consider the QoS of users and the allocation of network
service resource management and quality of service optimization. In Shi resources. To address this limitation, Wang et al. (2016) proposed a
et al. (2019), a cross-domain multi-layer cooperative network including scheme for joint resource allocation and service function chaining,
satellite, airspace platform and ground communication system is de- where they designed a cost model to strike a balance between net-
signed. Giambene et al. (2018) designed an architecture for future 5G work costs and service performance. In a similar vein, Mechtri et al.
networks utilizing SDN/NFV technology to converge satellite networks, (2016) put forward a novel eigen decomposition-based approach for
and proposes a network coding approach to divert satellite-terrestrial the placement of physical and virtual network function chains, with
traffic, which enables it to realize multi-path routing diversion in a a focus on maximizing providers’ profits through increased CPU and
satellite-terrestrial integrated environment. To solve the problem of bandwidth resource utilization. Shang et al. (2019) proposed a joint
resource allocation and management in 5G satellite network, a network SFC deployment and routing optimization problem. Since the problem
slicing management scheme based on SDN/NFV is proposed in Suzhi is NP-Hard, the author designs an approximate algorithm for candidate
et al. (2019) to meet the Qos requirements of different application path selection, which improves the efficiency of the solution. However,
scenarios and services. Zhang et al. (2017) proposed a hierarchical they did not take into account both the node mapping cost and the
network architecture to dynamically manage the shared resources by node instantiation cost, nor did they consider the aggregation scheme
using hierarchical controllers, which is based on the study of the of multiple VNFs. In recent years, reinforcement learning algorithms
vehicle-connected network application scenarios in the SAGIN, it also have gradually been applied in various fields by virtue of their inherent
looks forward to open research directions such as the provision of cus- advantages in solving decision-making problems. Pei et al. (2019)
tomized virtual network services, the use of software-defined networks proposed a VNF deployment algorithm based on dual-depth Q network,
for centralized control and management of heterogeneous resources, which utilizes a threshold-based strategy to deploy and release VNF
and how to protect software-defined network card and vehicle network instances. Simulation results demonstrate that the algorithm performs
security, but no in-depth discussion. well in terms of load balancing and throughput. Shah and Zhao (2020)
In the aspect of service resource management and optimization, proposed a deep reinforcement learning based SFC mapping algorithm
most of the existing researches adopt classical exact algorithm and to determine the deployment location of each VNF in the service chain
heuristic. In Varasteh et al. (2019), the problem is addressed by model- through appropriate reward design, state and action space formula-
ing it as a mixed integer linear programming problem. The model takes tions. However, when using intelligent learning algorithms to solve SFC
deployment problems in multi-party collaborative network scenarios
into account the diverse service and service resource requirements, as
such as SAGIN, the consideration of numerous feature variables also
well as the mobility of the space and air nodes. This paper proposes a
increases the complexity of state space solving (Wang et al., 2021).
method of deployment and routing of joint services based on mobile
Building upon the insights gained from previous research, we have
awareness, and simulates it in a small-scale environment using an
innovated and tailored our approach to the specific application sce-
optimization problem solver. Compared with the static routing method,
nario. With a focus on resource optimization and enhancing service
the cost of service and end-to-end delay are significantly reduced,
quality in SAGIN, this paper presents a novel SFC mapping algo-
but the efficiency of the solution in large-scale scenarios needs to be
rithm that incorporates virtual network function aggregation. This
improved. Wang et al. (2020) designed a reconfigurable service frame- approach aims to minimize mapping costs and enhance resource uti-
work based on SFC, and the VNF mapping and service routing problems lization through improved network service orchestration.
are considered in the sky-sky-earth network architecture, a heuristic
greedy algorithm with node characteristics and resource equalization is 3. Problem statement and system model
used to solve the problem. Li et al. (2018) proposed a framework for ar-
ranging the functional chain of transverse multi-domain services under This Section focuses on resource allocation for SFC requests in
the framework of satellite-ground cooperative network, a heuristic ser- SAGIN scenario, which requires the design of a topology for all SFC
vice function chain mapping algorithm is designed. Yang et al. (Zhang requests arriving in a longer period, i.e., the VNF nodes in the service

3
P. Zhang et al. Journal of Network and Computer Applications 224 (2024) 103829

Fig. 3. The underlying infrastructure.

physical nodes providing compute/storage/conversion resources, and

𝐸𝑠 denotes the set of all physical links providing bandwidth resources.
Fig. 2. Link Aggregation.
Assume that the total available resource capacity of each physical node
is 𝐶(𝑣𝑠 ), 𝑣𝑠 ∈ 𝑉𝑠 , 𝑐𝑜𝑠𝑡𝑣 (𝑣𝑠 ) denotes the mapping cost of the node; the
function chain are instantiated in different ways, while the designed total available bandwidth capacity of each physical link 𝐵(𝑒𝑠 ), where
topology is deployed in a reasonable manner to minimize the mapping 𝑒𝑠 ∈ 𝐸𝑠 . Also let 𝑟𝑉 (𝑣𝑠 ) denote the remaining available resources of
cost of the network service function chain and reduce user service the node and 𝑟𝐸 (𝑛𝑠 , 𝑚𝑠 ) denote the remaining available bandwidth of
latency while satisfying user service performance. the link, where 𝑛𝑠 , 𝑚𝑠 ∈ 𝑉𝑠 , 𝑙𝐸 denotes the link propagation delay and
𝑐𝑜𝑠𝑡𝐸 (𝑒𝑠 ) denotes the mapping cost of the link. As shown in Fig. 3, the
3.1. Description of the problem underlying physical topology has six physical nodes, represented by
circles, and a set of numbers in a rectangle next to each physical node,
We consider the general SAGIN topology as a hierarchical net- the left side indicates the computing resources owned by the node,
work architecture that includes space nodes (i.e., satellites), air nodes the right side indicates the remaining available resources, while the
(i.e., high-altitude platforms), and ground nodes (i.e., grounding de- physical network has seven links carrying communication between the
vices) (Zhang et al., 2022b). In the network topology on the right side nodes, the value next to the links on the left side is the total bandwidth
of Fig. 1, suppose that there are two service requests SFC-1 and SFC-2, capacity of the links, the right side is the remaining bandwidth size, in
the traffic needs to be in order through the VNF1-VNF2-VNF3, SFC-1 addition the triangle represents In addition, the triangle represents the
path through the ground nodes A-B-C-D to reach the end of the end to location of the system at the chain end of the service function.
complete the communication, SFC2 path through the space-air nodes E-
H-I to reach the end. It can be seen that when traffic congestion occurs 3.3. Service function chain requests
at the ground node, the SFC-2 path can complete the communication
and satisfy the service demand with the help of the air node, and Since one of the main types of networking that currently exists
significantly reduces the number of path hops. for SFC requests is serial connections, a user’s SFC request can be
Functional aggregation is the process of aggregating VNFs with the represented by a directed weighted graph 𝐺𝑟 = (𝑉𝑟 , 𝐸𝑟 ) in which the set
same functional type to form a VNF with greater resource requirements, of nodes 𝑉𝑟 denotes the virtual network function required by the current
and mapping the aggregated VNFs on the same physical server. To user, and the set of edges 𝐸𝑟 denotes the set of virtual links between
reduce the mapping cost, we can aggregate service function chains each virtual function.𝑓𝑔𝑖 denotes the function type of the 𝑖𝑡ℎ VNF in
𝑟
that have the same start and end points for simultaneous requests. the 𝑟𝑡ℎ service function chain, and 𝑟 denotes the resources required for
In practice, this means combining VNFs with the same function in the instantiation of a certain type of VNF. The annotation 𝑒𝑟 = (𝑛𝑟 , 𝑚𝑟 )
a service function chain to create a larger VNF that encompasses all ∈ 𝐸𝑟 denotes the connection relationship between VNF nodes 𝑛𝑟 and
resource requests. Importantly, the order of the VNFs within their 𝑚𝑟 . Also assume that 𝑣1𝑟 , 𝑣2𝑟 , … , 𝑣𝑙𝑟 , … , 𝑣𝐿
𝑟 denote the sequence of VNF
respective service function chains remains unchanged. nodes through which user traffic passes. Furthermore, let 𝑑𝑉 (𝑣𝑟 ), 𝑣𝑟
As shown in Fig. 2, there are two SFCs (SFC1 and SFC2), without ∈ 𝑉𝑟 denotes the resource requirements of the user VNF nodes; the
considering VNF instance sharing, SFC1 passes through A → C → E → annotation 𝑑𝐸 (𝑒𝑟 ) = 𝑑𝑒 (𝑛𝑟 , 𝑚𝑟 ) denotes the bandwidth requirements
F, and SFC2 passes through node A → B → D → F. It can be found between VNF nodes 𝑛𝑟 and 𝑚𝑟 . The annotation 𝑙𝑎𝑡(𝑎, 𝑎′ ) represents
that VNF1 exists in both SFCs, so we can reuse VNF1 instances, and the maximum tolerable service delay per SFC request. As shown in
deploy the VNF1 at node D to node C. Then SFC2 passes through Fig. 4, the SFC request consists of three sub-VNFs and two virtual
node A → B → D → C → E → F, which reduces the deployment links. The triangles represent the end systems of the SFC requests. The
cost. However, there will be some SFCs that map longer physical paths value inside the rectangle next to the node is the required computing
to reach shared VNF instances when deploying VNF instances, which capacity, the English character above the node represents the type of
increases the consumption of link bandwidth resources and violates function implemented by that VNF, the computing resources required
the latency requirements of SFCs, so how to balance the relationship to instantiate that function type are to its right, and the value above
between links and VNF instances is a problem that needs to be solved the virtual link between the nodes is its required bandwidth size.
at present. Functional aggregation is the aggregation of VNFs with the same
functional type to form a VNF with greater resource requirements, and
3.2. Physical network the mapping of the aggregated VNFs to the same physical server. To
further elaborate the meaning of functional aggregation, this section
In this paper, the underlying physical network is represented as assumes that the underlying physical network has a total of three
a weighted undirected graph 𝐺𝑠 = (𝑉𝑠 , 𝐸𝑠 ), 𝑉𝑠 denotes the set of all physical servers C1, C2, C3, as shown in Fig. 5, all with unit capacity

4
P. Zhang et al. Journal of Network and Computer Applications 224 (2024) 103829

Table 1
Definition of variables in the model.
Variable Definition
𝑣𝑗𝑆
𝑥𝑣𝑙 0-1 integer variable, 1 for the 𝑙 th virtual network
𝑟
function 𝑣𝑙𝑟 of request 𝑟 to the mapped physical node
𝑣𝑗𝑆 , otherwise 0
𝑖𝑓 ,𝑣𝑗 0-1 integer variable, service function type 𝑓 already
𝑆
instantiated on 𝑣𝑗𝑆 is 1, otherwise 0
Fig. 4. Virtual network requests. 𝑛 ,𝑚
0-1 integer variable, virtual link (𝑣𝑙𝑟 , 𝑣𝑙+1
𝑓𝑣𝑙𝑠,𝑣𝑙+1𝑠
𝑟 𝑟
𝑟 ) map is 1 on
physical link (𝑛𝑠 , 𝑚𝑠 ), otherwise 0

the SFC request topology, the aggregation of functions will cause a cor-
responding convergence of the physical links adjacent to them. If either
end of two virtual links in different requests is aggregated, such as
node 𝑉𝑟 and node 𝑉𝑟′ , the aggregated virtual network function node 𝑉𝑐
that requires more computing resources is formed. Then the adjacency
link of 𝑉𝑐 is the union of the adjacency links of the two VNF nodes,
that is, {(𝑣𝑟 , 𝑒)|∀𝑒 ∈ 𝑉𝑟 , (𝑣𝑟 , 𝑒) ∈ 𝐸𝑟 } ∪ {(𝑣𝑟′ , 𝑒)|∀𝑒 ∈ 𝑉𝑟′ , (𝑣𝑟′ , 𝑒) ∈ 𝐸𝑟′ }.
When both end nodes are aggregated, the two virtual links between
these two end nodes are also aggregated, and the aggregated virtual
link has a bandwidth capacity that equals the sum of the bandwidth
capacities of the two virtual links. Conversely when the two virtual
links are aggregated, the VNF nodes at each of their ends are also
Fig. 5. Function Aggregation. aggregated, and the computing resources required by the aggregated
virtual network function nodes are the sum of the computing resources
required by the VNF nodes at each end. Additionally, the required
1. There are two static service function chains in the system waiting instantiation resources remain the same as before aggregation.
for resource allocation, SFC1 and SFC2. SFC1 requests two virtual
network functions, VNF1 and VNF2, which require The instantiation 4. Problem formulation
resources for both VNF1 and VNF2 are 0.3. From Fig. 5(a), it can be
seen that in the one-to-one mapping strategy adopted for traditional This section utilizes optimization theory to mathematically formu-
virtual networks, these three VNF instances will be mapped to the nodes late the resource allocation problem based on functional aggregation as
C1, C2 and C3, respectively, since the instantiation cost of the nodes a Mixed Integer Linear Programming (MILP) problem. The objective is
is not taken into account. C1, C2 and C3, the three physical servers. to obtain a resource allocation solution that minimizes mapping costs.
However, here if the flexible resource allocation approach proposed Table 1 shows the decision variables used in the model:
by network virtualization technology is adopted, as in Fig. 5(b), the
VNF2 on server C3 is also deployed on server C2, so that only one 4.1. Virtual network function node mapping
instantiation of function 2 is used, reducing the resource mapping cost
by 0.3. To ensure that each VNF node in an accepted SFC request has and
When multiple SFC requests arrive, the network service provider can only be mapped to one physical node, we therefore introduce a
allocates sufficient network resources to the SFC requests based on node mapping constraint as shown in Eq. (1), which indicates that a
the resources required by the user and the required service perfor- VNF node cannot be instantiated to a different server with resource
mance, including the compute resources required by the VNF nodes, splitting.
the instantiation resources and the bandwidth resources required by ∑ 𝑗
𝑣𝑆
the virtual links. However, it is also necessary to take into account the ∀𝐺𝑟 , ∀𝑣𝑙𝑟 ∈ 𝑉𝑟 𝑥 =1 (1)
𝑣𝑙𝑟
𝑗
location of the end-system from which the user originates the request, 𝑣𝑆 ∈𝑉𝑆

so the network service provider needs to provide resource allocation Eq. (2) shows that the user initiating the SFC request has a defined
for multiple requests from one end-system at the same time, as well network access node and a request receiving point, that is an end-
as for requests from different end-systems, and also the order in which system pair, so 𝑚𝑎,𝑙𝑜𝑐(𝑎) = 1 in equation is not a decision variable:
traffic flows through the virtual network function nodes in the service
function chain. In this paper it is assumed that: 𝑃𝑉 ∶ 𝑉𝑟 → 𝑉𝑠 is
the set of physical nodes corresponding to the mapping of each VNF ∀𝑎 ∈ 𝐴, 𝑚𝑎,𝑙𝑜𝑐(𝑎) = 1 (2)
node in the user’s SFC request. Since this paper considers that VNF Eqs. (3) and (4) are complementary to each other and not only
nodes with the same type in different service requests can be mapped guarantee that if some function 𝑓 has been instantiated on server
on the same physical server, they are only instantiated once. If there 𝑣𝑗𝑆 , then there must be a VNF mapping of that type of functionality
is an instance of that VNF on the server, only the working resources requested by SFC to this server, and it also guarantees that as long as
of the VNF node need to be allocated and no further instantiation a VNF for a SFC request is mapped to server 𝑣𝑗𝑆 , then there must be
resources need to be allocated. Therefore, according to this feature, a corresponding function type 𝑓 already configured on this server: for
the VNF nodes in a given set of multiple SFC requests and the links ∀𝑓 ∈ 𝐹 , ∀𝑣𝑗𝑆 ∈ 𝑉𝑠 :
between them can be reasonably aggregated to form a new topology,
∑ ∑ 𝑗
𝑣𝑆
and the resources can then be allocated to the aggregated topology, 𝑥 ≤ 𝜔𝑖𝑓 ,𝑣𝑗 (3)
𝑣𝑙𝑟 𝑆
which can reduce the overhead of instantiating resources and thus 𝐺𝑟 𝑣𝑙 ∈𝑉 ,𝑓 𝑖 =𝑓
𝑟 𝑟 𝑔
𝑟
achieve the purpose of reducing the mapping cost. At the same time, ∑ ∑ 𝑗
𝑣𝑆
this Section only considers VNFs with different SFC requests to be 𝑖𝑓 ,𝑣𝑗 ≤ 𝑥 (4)
𝑆 𝑣𝑙𝑟
aggregated and mapped on the same server. Therefore, when designing 𝐺𝑟 𝑣𝑙 ∈𝑉 ,𝑓 𝑖 =𝑓
𝑟 𝑟 𝑔
𝑟

5
P. Zhang et al. Journal of Network and Computer Applications 224 (2024) 103829

Because in the traditional virtual network mapping strategy (Chowd- where 𝑇𝑛,𝑚 represents the deployment delay of each virtual link,
hury et al., 2011), VNF nodes belonging to the same SFC cannot and 𝛩 and 𝛤 are two adjustment factors. The equation shows that in
be mapped to the same physical node in order to ensure the high addition to the computational resource cost, the total node mapping
availability of users’ SFC requests, this Section also follows this design cost also has the cost of instantiating the virtual network function, so
principle: the algorithm designed in this paper is to minimize the instantiation
∑ 𝑗
𝑣𝑆 cost by placing VNF with the same function on the same server. At
∀𝐺𝑟 , ∀𝑣𝑗𝑆 ∈ 𝑉𝑆 𝑥 ≤1 (5) the same time, the virtual link mapping also has a mapping cost, so as
𝑣𝑙𝑟
𝑣𝑙𝑟 ∈𝑉𝑟 far as possible, the virtual link of the two end nodes to map the two
At the same time, each physical server needs to meet not only the closer to the two servers. Furthermore, we need to minimize the end-
computational resource requirements of each VNF node, but also the to-end latency of each virtual link. The end-to-end delay is calculated
instantiated resource requirements of the corresponding functional type by adding the transmission delay of each hop link in the SFC, assuming
of that VNF, and taking into account all the VNF nodes in the SFC that the transmission delay of each hop link in the SFC is 1 hop, the
request: delay mapped to the physical link is:
∑ ∑ 𝑗
𝑣𝑆 ∑ ∑ 𝐿−1
∑
∀𝑣𝑗𝑠 ∈ 𝑉𝑠 𝑥 ∗ 𝑑𝑉 (𝑣𝑟 ) 𝑇𝑛,𝑚 = 𝑓
𝑛𝑠 ,𝑚𝑠
𝑙𝐸 (𝑛𝑠 , 𝑚𝑠 )
𝑣𝑙𝑟
𝐺𝑟 𝑣𝑙 ∈𝑉 𝑣𝑙𝑟 ,𝑣𝑙+1
𝑟 (12)
𝑟 𝑟 (6) (𝑛𝑆 ,𝑚𝑆 )∈𝐸𝑆 𝐺𝑟 𝑙=1
∑
+ 𝑖𝑓 ,𝑣𝑗 ∗ 𝑟𝑓 ≤ 𝑟𝑉 (𝑣𝑗𝑆 ) ∀𝑚𝑠 .𝑛𝑠 ∈ 𝑉𝑠
𝑆
𝑓 ∈𝐹
The regulatory factors 𝛩 and 𝛤 are:
4.2. Virtual link mapping 1
𝛩= (13)
𝐶𝑚𝑎𝑥
To avoid closed loop paths or unnecessary extensions of the physical
1
mapping path during the link mapping phase, i.e., to ensure that only 𝛤 = (14)
𝑇𝑚𝑎𝑥
one of the physical links is mapped in one of the directions:
𝑛𝑠 ,𝑚𝑠 𝑚𝑠 ,𝑛𝑠 where 𝐶𝑚𝑎𝑥 is the maximum deployment cost and 𝑇𝑚𝑎𝑥 represents
𝑚𝑠 , 𝑛𝑠 ∈ 𝑉𝑠 , 𝑓 +𝑓 ≤1 (7) the delay threshold of the link.
𝑣𝑙𝑟 ,𝑣𝑙+1
𝑟 𝑣𝑙𝑟 ,𝑣𝑙+1
𝑟

To ensure that the flow conservation condition is satisfied for all

intermediate physical nodes in the physical path, except for the physical 5. Algorithm design
node mapped by the source virtual node after the virtual link maps to
the physical path: For the mixed integer programming model given in Section 3, the
∑ 𝑛 ,𝑚 𝑚 ,𝑛𝑠 𝑛 𝑛
SFC resource allocation problem can be reduced to VNEP (Chowdhury
(𝑓 𝑙𝑠 𝑙+1𝑠
− 𝑓 𝑙 𝑠 𝑙+1 ) = 𝑥 𝑠𝑙 − 𝑥 𝑠𝑙+1 et al., 2009). It is well known that the VNE problem is NP-hard, so the
𝑣𝑟 ,𝑣𝑟 𝑣𝑟 ,𝑣𝑟 𝑣𝑟 𝑣𝑟
𝑚𝑠 ∈𝑉𝑠 (8) SFC request deployment problem is naturally NP-hard as well. Under
𝑙 𝑙+1
𝑛𝑠 ∈ 𝑉𝑠 , ∀𝑣𝑟 , 𝑣𝑟 ∈ 𝑉𝑟 the condition of NP ≠ P, there is no polynomial time algorithm to solve
Furthermore, it is crucial to ensure that the underlying physical link the problem. In this Section, we propose Aggregation of Service Func-
has sufficient capacity to accommodate the bandwidth requirements tion Chain (A-SFC) algorithm for off-line resource deployment, which is
of the static SFC-requested virtual link during the mapping process. a heuristic algorithm that can effectively solve the SFC topology design
Therefore, it is necessary to satisfy the link capacity constraint: and mapping problem.

∑ 𝐿−1
∑ 𝑛𝑠 ,𝑚𝑠 5.1. Function aggregation-based service function chain mapping algorithm
𝑓 𝑑𝐸 (𝑣𝑙𝑟 , 𝑣𝑙+1
𝑟 ) ≤ 𝑟𝐸 (𝑛𝑠 , 𝑚𝑠 )
𝑣𝑙𝑟 ,𝑣𝑙+1
𝑟 (9)
𝐺𝑟 𝑙=1
In this section, we aim to address the challenge of service function
∀𝑚𝑠 .𝑛𝑠 ∈ 𝑉𝑠
chain mapping based on function aggregation. Before proceeding with
To ensure the QoS quality of user services, it is essential to ensure the mapping process, it is necessary to aggregate the collection of SFCs
that the total propagation delay of the physical path set mapped by the based on their functions. This involves combining the VNFs that have
virtual link is equal to or less than the delay required by the user. the same source and destination nodes into shareable VNFs. The main
∑
𝐿−1 objective of the service function chain aggregation is to determine the
𝑛𝑠 ,𝑚𝑠
𝑓 𝑙𝐸 (𝑛𝑠 , 𝑚𝑠 ) ≤ 𝑙𝑎𝑡(𝑎, 𝑎′ ), ∀𝐺𝑟 (10) topological order of the directed graph representing the SFC collection.
𝑣𝑙𝑟 ,𝑣𝑙+1
𝑟
𝑙=1 To achieve this, we assume that there exists a sequential order of
4.3. Objective function VNFs from node A to node B within each SFC. In the topological
graph, this sequential order implies an edge from node A to node B.
To minimize the mapping cost of the service function chain and In simpler terms, satisfying the condition of a function aggregation
reduce service delay, this paper finds a suitable physical node mapping requires ensuring that node A appears before node B in the entire
for each VNF node in the SFC request, and find the best physical path topological graph. In order to implement the SFC aggregation and
which can be mapped for its adjacent virtual link. Thus the objective satisfy the mentioned conditions, we follow the approach described in
function of the optimization model can therefore be shown in Eq. (11). the literature (Kahn, 1962). We adopt the search mapping algorithm
based on isomorphic graph, vnmFlib algorithm, as mentioned in the
∑ ∑ ∑ 𝑣𝐽𝑆
literature (Lischka and Karl, 2009), as the fundamental framework for
𝑚𝑖𝑛(𝛩( ( 𝑥 × 𝑑𝑉 (𝑣𝑙𝑟 ) SFC. The virtual link is isomorphic to a physical path as described
𝑣𝑙𝑟
𝑗 𝐺𝑟 𝑣𝑙 ∈𝑉
𝑣𝑆 ∈𝑉𝑆 𝑟 𝑟 in the algorithm means that a physical link is found for that virtual
∑ link that satisfies both the bandwidth requirements and the hop count
+ 𝑖𝑓 ,𝑣𝑗 × 𝑟𝑓 ) × 𝑐𝑜𝑠𝑡𝑉 (𝑣𝑗𝑆 )
𝑓 ∈𝐹 𝑆
(11) constraints set by the algorithm.
Since the link aggregation process may increase the consumption
∑ ∑ 𝐿−1
∑ 𝑛𝑆 ,𝑚𝑆
+ 𝑓 𝑑𝐸 (𝑣𝑙𝑟 , 𝑣𝑙+1
𝑟 ) × 𝑐𝑜𝑠𝑡𝐸 (𝑒𝑆 ))
of bandwidth resources while reducing the functional configuration
𝑣𝑙𝑟 ,𝑣𝑙+1
𝑟
(𝑛𝑆 ,𝑚𝑆 )∈𝐸𝑆 𝐺𝑟 𝑙=1 overhead of VNF nodes, and in order to shorten the algorithm runtime,
+ 𝛤 𝑇𝑛,𝑚 ) this Section compromises the instantiated mapping cost and the link

6
P. Zhang et al. Journal of Network and Computer Applications 224 (2024) 103829

mapping cost. The upper bound of link aggregation can be derived from
the following:
∑
𝑓 𝑟𝑓
𝑐𝑜𝑠𝑡𝐸 (𝑒𝑆 ) ∗ 𝐷 ≤ 2 ∗ 𝑐𝑜𝑠𝑡𝑣 (𝑣𝑆 ) ∗ (15)
𝑠𝑖𝑧𝑒(𝑓 )
The implication of this formula is that the reduction of instantiation
mapping cost after the link aggregation is greater than the increase of Fig. 6. SFC Delay Estimation.
the bandwidth demand after link aggregation. The upper bound of link
aggregation can therefore be obtained as follows:
∑
𝑓 𝑟𝑓 6.1. Setting of experimental environment
2 ∗ 𝑐𝑜𝑠𝑡𝑉 (𝑣𝑠 ) ∗ 𝑠𝑖𝑧𝑒(𝑓 ) (16)
𝐷𝑈 =
𝑐𝑜𝑠𝑡𝐸 (𝑒𝑠 ) In this paper, we analyze the simulation results using a network
The lower bound value 𝐷𝐿 is defined as the average virtual link topology with 58 nodes and 87 links(58–87 topology). The network
bandwidth requirement. It is also necessary to consider the computing topology is derived from a study by (Knight et al., 2011) that was
resource requirements and the instantiation of functions after end-node dedicated to collecting information about different network topologies
aggregation when link aggregation. Hence the need to satisfy: in different countries. In addition, to validate the generality and effec-
tiveness of the algorithm, we also used the GT-ITM tool to generate a
𝑟(𝑓𝑔𝑙 ) + 𝑑𝑣 (𝑣𝑙𝑟 ) + 𝑑𝑣 (𝑣𝑙𝑟′ ) ≤ 𝑟𝑣 (𝑣𝑖𝑠 ) (17)
𝑟 topology including 30 nodes and 59 links (30–59 topology) for com-
The optimization objective of the current section is to minimize parison. The computing resources and link bandwidth resources of the
the cost of mapping, so all SFC requests are topologically designed physical node are both 10,000. In order to verify the cost optimization
first, that is, 𝑘 aggregation schemes are generated according to the brought by the algorithm, we assume that the unit cost of bandwidth
interval given in above. Each scheme corresponds to several groups of resources and node computing resources for each link is 3 (Luizelli
new topologies formed by link aggregation in clusters, and these new et al., 2015).
topologies are mapped to the underlying physical facilities one group In the case of a single time slot, there are 4–8 VNFs per SFC
at a time by using vnmlib algorithm. The details of the A-SFC algorithm request (Zhang et al., 2022b), the VNF function type in one SFC
are listed below: request is different, the resources required for each VNF are uniformly
Algorithm 1 Aggregation of Service Function Chain (A-SFC) Algorithm distributed [50, 100], and each SFC request is a chain structure; the
Input: SFC request: 𝐺𝑟 = (𝑉𝑟 , 𝐸𝑟 ); physical network: 𝐺𝑠 = (𝑉𝑠 , 𝐸𝑠 ); bandwidth requirements of the virtual links follow a uniform distribu-
endpoint pairs: 𝐴𝑝𝑎𝑖𝑟𝑠 (𝑎, 𝑎′ ); tion within the range of [50,100]. Similarly, the instantiated resources
Output: VNF mapping: 𝑃𝑉 (𝑣𝑟 )|∀𝑣𝑟 ∈ 𝑉𝑟 ; virtual link mapping: of each function also adhere to a uniform distribution within the range
𝑃𝐸 (𝑒𝑟 )|∀𝑒𝑟 ∈ 𝐸𝑟 ; minimum mapping costs; of [200, 400], and each SFC request has a fixed end-system access
1: Let 𝐷𝐿 be equal to 𝑏𝑜𝑢𝑛𝑑; simulation experimental topology.
2: Calculate 𝐷𝑈 according to the equation (13); The primary objective of this paper is to minimize the cost associ-
3: if 𝐷𝐿 > 𝐷𝑈 then ated with mapping the service function chain, while minimizing the
4: 𝐷𝐿 and 𝐷𝑈 exchange numerical values; latency associated with SFC requests. Therefore, this section focuses
5: end if on the mapping cost of A-SFC algorithm and other algorithms with
6: while 𝐷𝐿 ≤ 𝑏𝑜𝑢𝑛𝑑 ≤ 𝐷𝑈 do different number of SFC requests, including link forwarding bandwidth
7: Put the virtual links of the same source node into the collection; cost and node deployment cost, and also compares the service delay
8: Perform link aggregation by the method for obtaining topolog- of SFC requests under different algorithms. Then finally this section
ical order; analyzes and discusses the utilization of VNF and the acceptance rate
9: Use the vnmlib algorithm to map the aggregation topology; of service requests.
10: Add up the 𝑏𝑜𝑢𝑛𝑑 and average bandwidth requirements of the
virtual links; 6.2. Simulation performance index
11: end while
12: return the best mapping plan;
To validate the effectiveness of the algorithm, this section will
utilize four specific performance indicators as the objects of simulation
The A-SFC designed in this paper is based on clustering and then analysis.
mapping, and the mapping process adopts the way of node and link
coupling mapping, therefore, the time complexity of the algorithm • Mapping cost: the total cost of node mapping and link mapping.
will be increased if the delay requirement is strongly constrained As shown in Eq. (11).
in the mapping stage. However, in order to satisfy the user’s delay • Average delay: the average of all static SFC request delays, which
requirement, two attribute values, 𝑝𝐹 𝑖𝑟𝑠𝑡 and 𝑝𝐸𝑛𝑑, are defined for is used to measure the performance of the service.
each aggregated VNF node, which are the number of hops (Yang et al., • VNF utilization: actual VNF instance utilization as a percentage
2018) to the start of the request and the number of hops to the of total VNF instances required.
receiver of the request, both hops have an initial value of −1, and • Acceptance rate: percentage of mappings that can be successfully
both attribute values are updated after the mapping node is solved. As accepted when the number of SFCs is increased.
shown in Fig. 6, during resource allocation to VNF node C, only node
A in this SFC topology has been mapped to the physical server P1, so Due to the lack of relevant standardized data, different strategy
𝑝𝐹 𝑖𝑟𝑠𝑡𝐶 = 𝑝𝐹 𝑖𝑟𝑠𝑡𝐴 + |𝑃 𝑎𝑡ℎ(𝑃 1, 𝐶)|, 𝑝𝐸𝑛𝑑𝐶 = |𝑃 𝑎𝑡ℎ(𝑃 2, 𝑆2)|. In addition, objectives, the existing theoretical literature also suffers from the dif-
𝑝𝐹 𝑖𝑟𝑠𝑡𝐶 +𝑝𝐸𝑛𝑑𝐶 is used to roughly estimate the request delay of node C ficulty of cross-sectional comparison of methods. To validate the pro-
mapping in the physical device. If the upper delay limit is exceeded, the posed algorithm’s effectiveness, we have selected the following service
physical (P2) node is not considered as an alternative mapping node. function chain mapping algorithms for comparison:

6. Performance evaluation 1) Layered Graph (LG) algorithm (Yin and Ma, 2018). The classical
heuristic algorithm is utilized to convert the configuration of
In this section, we begin by explaining the simulation setup and sub- application sessions requiring intermediate processing into a
sequently compare its performance with several different algorithms. more manageable conventional shortest path algorithm.

7
P. Zhang et al. Journal of Network and Computer Applications 224 (2024) 103829

Fig. 7. Bandwidth cost for different number of SFC requests.

2) Shortest Path Heuristic (SPH) algorithm (Zhang et al., 2021).

Fig. 8. Deployment Costs for Different Number of SFC Requests.
The algorithm consistently searches for the nodes with the short-
est delay in order to map the VNF.
3) Greedy on Used Server (GUS) (Zhang et al., 2021). It priori- illustrates a comparison of these algorithms specifically focusing on
tizes placement in shareable VNFs to avoid taking up additional their impact on deployment costs. The SPH algorithm has the highest
unused servers. deployment cost, which is higher than this paper’s algorithm 206.5 on
average, because each SFC is greedily deployed on the shortest path
and ignores shared VNF instances. Similarly, the LG algorithm is a
6.3. Analysis of the simulation results heuristic algorithm that utilizes the Dijkstra algorithm to obtain paths.
However, it does not consider the possibility of sharing VNF instances.
In this section, a variable F is defined that represents is the num- As a result, the average deployment cost of the LG algorithm is 133.25
ber of VNF function types, the value is 15 to 20, for more accurate higher than that of the A-SFC algorithm proposed in this paper. By
simulation results, we take as our simulation the average of the cost comparison, the A-SFC algorithm in this paper well balances the link
of mapping the number of function types from 15 to 20. This section forwarding cost and deployment cost. A-SFC algorithm aggregates the
defines the number of SFCs as follows [10,20,30,40]. functionally identical nodes in the service function chains with the
Fig. 7 illustrates a comparison of these algorithms specifically fo- same end node pairs, and then places the new topology formed after
cusing on their impact on bandwidth cost. The average bandwidth aggregation onto the physical topology. Since the mapping process
cost of GUS algorithm is higher than that of A-SFC algorithm 202.25. is performed for the entire chain with the same source destination
In order to share VNF instances as much as possible, the algorithm nodes, the impact of the sequential mapping of SFC requests is reduced
inevitably increases the path length. Furthermore, it is apparent that to a certain extent, which reduces the mapping cost. Meanwhile, the
the overall bandwidth cost in the 58–87 topology is lower than that in A-SFC algorithm aggregates the virtual links with the same source
the 30–59 topology. This is because the 58–87 topology is relatively catalog nodes by different aggregation upper limit values, and then
dense compared to the 30–59 topology. Consequently, there are more maps the aggregated new topology by the correlation function value,
opportunities to select physical paths with fewer hops during the link thus reducing the mapping cost.
mapping process, resulting in a reduction in bandwidth costs, which Fig. 9 illustrates the performance of the algorithms by measuring the
is also the reason for the lower average latency of the links. Fig. 8 link average delay. It is worth mentioning that the SPH algorithm shows

8
P. Zhang et al. Journal of Network and Computer Applications 224 (2024) 103829

Fig. 10. VNF utilization with different number of SFC requests.

Fig. 9. Average latency with different number of SFC requests.

deployment cost and thus avoids excessive increase in bandwidth cost
due to link forwarding. Such consistency applies to different underlying
a good advantage in reducing the response latency, which is on average physical networks.
2 hops lower than the A-SFC algorithm, because SPH aims to reduce
Fig. 11 illustrates the comparison of request acceptance rates for dif-
end-to-end response latency by integrating a maximum number of VNFs
ferent algorithms as the number of SFC requests increases. As depicted
into the topology nodes. However, the figure shows that the A-SFC algo-
in the figure, when the number of SFCs increases, the request accep-
rithm also performs well in terms of average delay. This is because the
tance rate of this paper’s service function chain deployment algorithm
A-SFC algorithm utilizes a service function chain collection approach
based on functional aggregation is higher than that of other algorithms.
for mapping, which mitigates the impact of sequentially mapping ser-
e.g., when the number of SFCs reaches 1000, the request acceptance
vice function chains. As a result, it reduces mapping cost and minimizes
rates of SPH and LG are 61.1% and 61.52%, respectively, the request
the average physical link collection mapped by virtual links, leading to
acceptance rate of GUS is 52.68%, and the service acceptance rate of
reduced request latency. Furthermore, the A-SFC algorithm reduces the
this paper’s algorithm is 65.12%. Because SPH prioritizes the shortest
number of iterative mappings and reduces the delay estimation error by
forming VNF nodes with greater resource requirements through intra- path deployment, it leads to the existence of VNF instances in the
aggregation. Likewise, the variation in sparsity between the 58–87 and network that exclusively occupy the server resources. LG will select
30–59 topologies leads to a lower overall average latency in the 58–87 the node of the next hop by hierarchically obtaining through Dijkstra
topology compared to the 30–59 topology. algorithm, which saves the bandwidth resources but inevitably results
Fig. 10 provides a comparison of VNF utilization between the A-SFC in the increase of the path length, which makes the link bandwidth
algorithm and other algorithms for different numbers of SFC request resources unable to satisfy the subsequent SFCs. GUS, although it shares
entries. As shown in the figure, since GUS prioritizes VNF instance the VNF instances, similarly increases the path length, thus causing the
sharing, it has the highest VNF instance utilization rate, which is higher same result as the LG algorithm. Therefore, the proposed algorithm can
than that of this paper’s algorithm by 22.42%, while SPH and LG support more SFCs in the network. Both network topologies show the
prioritize shortest-path deployment and consider less VNF instance same trend.
sharing, so their utilization rates are lower than that of this paper’s In summary, the A-SFC algorithm reduces the cost of service func-
algorithm, which are lower than that of this paper’s algorithm by tion chain mapping compared to other methods by aggregating links
8.14% and 18.8%, respectively. Although the utilization rate in this within the same end system and nodes with similar functions. By
paper is lower than that of GUS, it sacrifices part of the VNF instance forming all arriving SFC requests into a collection and establishing

9
P. Zhang et al. Journal of Network and Computer Applications 224 (2024) 103829

that the A-SFC algorithm outperforms other algorithms in terms of

reducing delay and mapping cost.

CRediT authorship contribution statement

Peiying Zhang: Conceptualization, Investigation, Methodology,

Software, Visualization, Writing – original draft. Kunkun Yan:
Conceptualization, Methodology, Resources, Visualization, Writing –
review & editing. Neeraj Kumar: Conceptualization, Funding acqui-
sition, Methodology, Project administration, Resources, Supervision,
Visualization, Writing – review & editing. Lizhuang Tan: Con-
ceptualization, Methodology, Resources, Supervision, Visualization,
Writing – review & editing. Mohsen Guizani: Conceptualization,
Supervision, Visualization, Writing – review & editing. Konstantin
Igorevich Kostromitin: Conceptualization, Methodology, Project
administration, Resources, Writing – review & editing. Jian Wang:
Investigation, Project administration, Visualization. Jianyong Zhang:
Data curation, Formal analysis, Validation.

Declaration of competing interest

The authors declare that they have no known competing financial

interests or personal relationships that could have appeared to
influence the work reported in this paper.

Data availability

Data will be made available on request.

Acknowledgments

This work is partially supported by the Natural Science Foundation

of Shandong Province under Grant ZR2023LZH017, ZR2022LZH015,
and 2023QF025 (for experimental data collection), partially supported
by the National Natural Science Foundation of China under Grant
Fig. 11. Acceptance rate of service requests with different number of SFC requests. 62173345 (for experimental data administration), partially supported
by the open project of the Key Laboratory of All Optical Network
connection relationships between them, it enables node and link co- and Advanced Telecommunication Network of EMC, Ministry of Educa-
mapping to consider more neighboring resources in a comprehensive tion, Beijing Jiaotong University under Grant AON2023K01 (for prob-
manner, resulting in more efficient resource allocation. This approach lem investigation), partially supported by the Integrated Innovation
also reduces VNF instantiation costs and overall resource consumption, of Science, Education and Industry of Qilu University of Technology
making it an effective and cost-efficient solution for SFC provisioning. (Shandong Academy of Sciences) under Grant 2023PX057 (for model
At the same time, the node mapping process reduces the request latency development), partially supported by the Talent Project of Qilu Uni-
to a certain extent by approximating its latency size as a parameter for versity of Technology (Shandong Academy of Sciences) under Grant
selecting alternative physical servers. 2023RCKY141 (for model verification), partially supported by the Open
Foundation of State Key Laboratory of Integrated Services Networks
7. Conclusion (Xidian University) under Grant ISN23-09 (for comparative experiment
and research), partially supported by the RSF project under Grant
In order to effectively improve the efficiency of service resource 22-71-10095 (for checking metrics), partially supported by the Fun-
management and optimization in SAGIN, this paper investigates the damental Research Funds for the Central Universities under Grant
dynamic orchestration problem of SFC, and designs a SFC orchestration 2023JBZY014 (for ablation experiment).
architecture and system model of the SAGIN network. Under this
architecture, the problem is modeled as a MILP model with resource References
and service request as constraints, and minimization of mapping cost
and end-to-end delay as optimization objectives. Finally, this paper Alzahrani, Bander, Oubbati, Omar Sami, Barnawi, Ahmed, Atiquzzaman, Mo-
proposes A-SFC, an innovative algorithm based on a functional aggre- hammed, Alghazzawi, Daniyal, 2020. UAV assistance paradigm: State-of-the-art in
applications and challenges. J. Netw. Comput. Appl. 166, 102706.
gation. Initially, the nodes and links in the SFC request topology are
Chen, Jia, Chen, Jing, Guo, Kuo, 2023. Queue-aware service orchestration and adaptive
reasonably aggregated, i.e., the SFC requests with the same endpoints parallel traffic scheduling optimization in SDNFV-enabled cloud computing. IEEE
are grouped into a collection. Then the virtual links with the same Trans. Cloud Comput. 1–17.
source destination nodes in the collection are aggregated according Chowdhury, NM Mosharaf Kabir, Rahman, Muntasir Raihan, Boutaba, Raouf, 2009.
Virtual network embedding with coordinated node and link mapping. In: IEEE
to different aggregation bandwidth capacities. Afterwards, they are
INFOCOM 2009. IEEE, pp. 783–791.
mapped to the same physical servers to join the SFC deployment and Chowdhury, Mosharaf, Rahman, Muntasir Raihan, Boutaba, Raouf, 2011. Vineyard:
path selection problems in order to save the link forwarding cost and Virtual network embedding algorithms with coordinated node and link mapping.
efficiently reuse the VNF instances. The simulation results demonstrate IEEE/ACM Trans. Netw. 20 (1), 206–219.

10
P. Zhang et al. Journal of Network and Computer Applications 224 (2024) 103829

Eramo, Vincenzo, Miucci, Emanuele, Ammar, Mostafa, Lavacca, Francesco Giacinto, Wang, Guangchao, Zhou, Sheng, Zhang, Shan, Niu, Zhisheng, Shen, Xuemin, 2020. SFC-
2017. An approach for service function chain routing and virtual function network based service provisioning for reconfigurable space-air-ground integrated networks.
instance migration in network function virtualization architectures. IEEE/ACM IEEE J. Sel. Areas Commun. 38 (7), 1478–1489.
Trans. Netw. 25 (4), 2008–2025. Yang, Binxu, Chai, Wei Koong, Xu, Zichuan, Katsaros, Konstantinos V, Pavlou, George,
Giambene, Giovanni, Kota, Sastri, Pillai, Prashant, 2018. Satellite-5G integration: A 2018. Cost-efficient NFV-enabled mobile edge-cloud for low latency mobile
network perspective. Ieee Netw. 32 (5), 25–31. applications. IEEE Trans. Netw. Serv. Manag. 15 (1), 475–488.
Gu, Shushi, Zhang, Qinyu, Xiang, Wei, 2020. Coded storage-and-computation: A new Yang, Song, Li, Fan, Trajanovski, Stojan, Yahyapour, Ramin, Fu, Xiaoming, 2020. Recent
paradigm to enhancing intelligent services in space-air-ground integrated networks. advances of resource allocation in network function virtualization. IEEE Trans.
IEEE Wirel. Commun. 27 (6), 44–51. Parallel Distrib. Syst. 32 (2), 295–314.
Han, Bo, Gopalakrishnan, Vijay, Ji, Lusheng, Lee, Seungjoon, 2015. Network function Yin, Xianyong, Ma, Yan, 2018. Aggregation service function chain mapping plan based
virtualization: Challenges and opportunities for innovations. IEEE Commun. Mag. on beetle antennae search algorithm. In: Proceedings of the 2nd International
53 (2), 90–97. Conference on Telecommunications and Communication Engineering. pp. 225–230.
Kahn, Arthur B., 1962. Topological sorting of large networks. Commun. ACM 5 (11), Zhang, Qixia, Liu, Fangming, Zeng, Chaobing, 2021. Online adaptive interference-aware
558–562. VNF deployment and migration for 5G network slice. IEEE/ACM Trans. Netw. 29
Knight, Simon, Nguyen, Hung X, Falkner, Nickolas, Bowden, Rhys, Roughan, Matthew, (5), 2115–2128.
Zhang, Peiying, Yang, Pan, Kumar, Neeraj, Guizani, Mohsen, 2022a. Space-air-ground
2011. The internet topology zoo. IEEE J. Sel. Areas Commun. 29 (9), 1765–1775.
Kua, Jonathan, Loke, Seng W, Arora, Chetan, Fernando, Niroshinie, integrated network resource allocation based on service function chain. IEEE Trans.
Ranaweera, Chathurika, 2021. Internet of things in space: A review of opportunities Veh. Technol. 71 (7), 7730–7738.
Zhang, Peiying, Zhang, Yi, Kumar, Neeraj, Guizani, Mohsen, 2022b. Dynamic sfc
and challenges from satellite-aided computing to digitally-enhanced space living.
embedding algorithm assisted by federated learning in space-air-ground integrated
Sensors 21 (23), 8117.
network resource allocation scenario. IEEE Internet Things J..
Li, Guanglei, Zhou, Huachun, Feng, Bohao, Li, Guanwen, Xu, Qi, 2018. Horizontal-
Zhang, Ning, Zhang, Shan, Yang, Peng, Alhussein, Omar, Zhuang, Weihua,
based orchestration for multi-domain SFC in SDN/NFV-enabled satellite/terrestrial
Shen, Xuemin Sherman, 2017. Software defined space-air-ground integrated
networks. China Commun. 15 (5), 77–91.
vehicular networks: Challenges and solutions. IEEE Commun. Mag. 55 (7), 101–109.
Lischka, Jens, Karl, Holger, 2009. A virtual network mapping algorithm based on
subgraph isomorphism detection. In: Proceedings of the 1st ACM Workshop on
Virtualized Infrastructure Systems and Architectures. pp. 81–88. Peiying Zhang is currently an Associate Professor with the
Luizelli, Marcelo Caggiani, Bays, Leonardo Richter, Buriol, Luciana Salete, Barcel- College of Computer Science and Technology, China Uni-
los, Marinho Pilla, Gaspary, Luciano Paschoal, 2015. Piecing together the NFV versity of Petroleum (East China). He received his Ph.D. in
provisioning puzzle: Efficient placement and chaining of virtual network functions. the School of Information and Communication Engineering
In: 2015 IFIP/IEEE International Symposium on Integrated Network Management. at University of Beijing University of Posts and Telecom-
IM, IEEE, pp. 98–106. munications in 2019. He has published multiple IEEE/ACM
Mauro, Mario Di, Galatro, Giovanni, Postiglione, Fabio, Tambasco, Marco, 2022. Trans./Journal/Magazine papers since 2016, such as IEEE
Performability of network service chains: Stochastic modeling and assessment of TII, IEEE T-ITS, IEEE TVT, IEEE TNSE, IEEE TNSM, IEEE
softwarized IP multimedia subsystem. IEEE Trans. Dependable Secure Comput. 19 TETC, IEEE Network and etc. He served as the Technical
(5), 3071–3086. Program Committee of AAAI’24, AAAI’23, IEEE ICC’23, IEEE
Mechtri, Marouen, Ghribi, Chaima, Zeghlache, Djamal, 2016. A scalable algorithm for ICC’22, and INFOCOM Wireless-Sec 2023. He is the Leading
the placement of service function chains. IEEE Trans. Netw. Serv. Manag. 13 (3), Guest Editor of Drones, Mathematics, Electronics, Wireless
533–546. Communications and Mobile Computing, and etc. He is
Nguyen, Minh, Dolati, Mahdi, Ghaderi, Majid, 2020. Deadline-aware SFC orchestration the editorial board of Drones, CMC-Computers, Materials &
under demand uncertainty. IEEE Trans. Netw. Serv. Manag. 17 (4), 2275–2290. Continua, Mobile Information Systems, International Journal
Pei, Jianing, Hong, Peilin, Pan, Miao, Liu, Jiangqing, Zhou, Jingsong, 2019. Optimal of Computational Intelligence Systems and Artificial Intelli-
VNF placement via deep reinforcement learning in SDN/NFV-enabled networks. gence and Applications (AIA). His research interests include
IEEE J. Sel. Areas Commun. 38 (2), 263–278. semantic computing, future internet architecture, network
Shafique, Kinza, Khawaja, Bilal A., Sabir, Farah, Qazi, Sameer, Mustaqim, Muhammad,
virtualization, and artificial intelligence for networking.
2020. Internet of Things (IoT) for next-generation smart systems: A review of
current challenges, future trends and prospects for emerging 5G-IoT scenarios. IEEE
Access 8, 23022–23040. Kunkun Yan is currently pursuing the graduate degree with
Shah, Hurmat Ali, Zhao, Lian, 2020. Multiagent deep-reinforcement-learning-based the College of Computer Science and Technology, China
virtual resource allocation through network function virtualization in Internet of University of Petroleum (East China). He research interests
Things. IEEE Internet Things J. 8 (5), 3410–3421. include network function virtualization and service function
Shang, Xiaojun, Liu, Zhenhua, Yang, Yuanyuan, 2019. Network congestion-aware online chain mapping.
service function chain placement and load balancing. In: Proceedings of the 48th
International Conference on Parallel Processing. pp. 1–10.
Sheng, Jie, Cai, Xingqiang, Li, Qingyang, Wu, Cheng, Ai, Bo, Wang, Yiming,
Kadoch, Michel, Yu, Peng, 2021. Space-air-ground integrated network development
and applications in high-speed railways: A survey. IEEE Trans. Intell. Transp. Syst.
23 (8), 10066–10085.
Shi, Yongpeng, Cao, Yurui, Liu, Jiajia, Kato, Nei, 2019. A cross-domain SDN architecture
for multi-layered space-terrestrial integrated networks. IEEE Netw. 33 (1), 29–35. Neeraj Kumar (Senior Member, IEEE) is currently working
Sun, Gang, Zhou, Run, Sun, Jian, Yu, Hongfang, Vasilakos, Athanasios V, 2020. as a Full Professor with the Department of Computer Sci-
Energy-efficient provisioning for service function chains to support delay-sensitive ence and Engineering, Thapar Institute of Engineering and
applications in network function virtualization. IEEE Internet Things J. 7 (7), Technology, Patiala, India. He has published more than 400
6116–6131. technical research papers, which are cited more than 43637
Suzhi, Cao, Junyong, Wei, Hao, Han, Yi, Zhao, Shuling, Yang, Lei, Yan, Shaojun, Wu, times by researchers across the globe with a current H-
Yongsheng, Gong, 2019. Space edge cloud enabling network slicing for 5G index of 114. His broad research areas are green computing
satellite network. In: 2019 15th International Wireless Communications & Mobile and network management, IoT, big data analytics, deep
Computing Conference. IWCMC, IEEE, pp. 787–792. learning, and cyber-security. He has also edited/authored
Van Bemten, Amaury, Guck, Jochen W, Vizarreta, Petra, Machuca, Carmen Mas, ten books with international/national publishers. He is a
Kellerer, Wolfgang, 2018. LARAC-SN and mole in the hole: Enabling routing highly-cited researcher from WoS from 2019 to 2021. Prof.
through service function chains. In: 2018 4th IEEE Conference on Network Kumar has organized various special issues of journals of
Softwarization and Workshops. NetSoft, IEEE, pp. 298–302. repute from IEEE, Elsevier, and Springer. He has been the
Varasteh, Amir, Hofmann, Sandra, Deric, Nemanja, He, Mu, Schupke, Dominic, Workshop Chair at IEEE GLOBECOM 2018, IEEE INFOCOM
Kellerer, Wolfgang, Machuca, Carmen Mas, 2019. Mobility-aware joint service 2020, and IEEE ICC 2020, and the track chair. He is serving
placement and routing in space-air-ground integrated networks. In: ICC 2019-2019 as an Editor of ACM Computing Surveys, IEEE Transactions
IEEE International Conference on Communications. ICC, IEEE, pp. 1–7. on Services Computing, IEEE Transactions on Network and
Wang, Chao, Liu, Lei, Jiang, Chunxiao, Wang, Shangguang, Zhang, Peiying, Shen, Shi- Service Management, IEEE Network Magazine, IEEE Com-
gen, 2021. Incorporating distributed DRL into storage resource optimization of munications Magazine, Journal of Networks and Computer
space-air-ground integrated wireless communication network. IEEE J. Sel. Top. Applications (Elsevier), Computer Communications (Else-
Sign. Proces. 16 (3), 434–446. vier), and International Journal of Communication Systems
Wang, Luhan, Lu, Zhaoming, Wen, Xiangming, Knopp, Raymond, Gupta, Rohit, 2016. (Wiley).
Joint optimization of service function chaining and resource allocation in network
function virtualization. IEEE Access 4, 8084–8094.

11
P. Zhang et al. Journal of Network and Computer Applications 224 (2024) 103829

Lizhuang Tan is currently an Assistant Professor with

Shandong Provincial Key Laboratory of Computer Networks, Kostromitin Konstantin Igorevich received the gradua-
Shandong Computer Science Center (National Supercom- tion degree (Hons.) from the Municipal Secondary School,
puter Center in Jinan), Qilu University of Technology Chelaybinsk, Russia, and joined the Physical Department
(Shandong Academy of Sciences). He received his Ph.D. de- of Education, Chelaybinsk State University, Chelyabinsk,
gree from School of Electronic and Information Engineering, with specialization in physics, the bachelor’s and master’s
Beijing Jiaotong University in 2022. He has published more degrees in physics with specialization in chair physics
than 20 journal or conference papers, such as USENIX NSDI, of condensed matter from Chelyabinsk State University,
IEEE TNSM, ELSEVIER CN, APNOMS, etc. His research Chelyabinsk, Russia, in 2008 and 2010, respectively, and the
interests include network measurement, management and Ph.D. degree in physics from Chelyabinsk State University,
optimization, especially software-defined networking and in 2013, with thesis titled Researching of Magnetocaloric
data center networking. Effect in Antiferromagnetics and Twins Moving in Heusler
alloys with specialization in physics of condensed matter in
the Dissertation Council.
Mohsen Guizani (Fellow, IEEE) received the BS (with dis-
tinction), MS and Ph. D. degrees in Electrical and Computer
engineering from Syracuse University, Syracuse, NY, USA Jian Wang is currently a Professor and serves as the Head
in 1985, 1987 and 1990, respectively. He is currently a of the Laboratory for Intelligent Information Processing with
Professor of Machine Learning and the Associate Provost the College of Science, China University of Petroleum (East
at Mohamed Bin Zayed University of Artificial Intelligence China). His research interests include computational intelli-
(MBZUAI), Abu Dhabi, UAE. Previously, he worked in gence, differential programming, clustering, fuzzy systems,
different institutions in the USA. His research interests evolutionary computation. Prof. Wang serves as an Associate
include applied machine learning and artificial intelligence, Editor for the IEEE Transactions on Neural Networks and
Internet of Things (IoT), intelligent autonomous systems, Learning Systems, IEEE Transactions on Emerging Topics
smart city, and cybersecurity. He was elevated to the IEEE in Computational Intelligence, Information Sciences, Inter-
Fellow in 2009 and was listed as a Clarivate Analytics national Journal of Machine Learning and Cybernetics, and
Highly Cited Researcher in Computer Science in 2019, Journal of Applied Computer Science Methods. He also
2020, 2021 and 2022. Dr. Guizani has won several research serves on the Editorial Board for the Neural Computing \&
awards including the “2015 IEEE Communications Society Applications and Complex \& Intelligent Systems.
Best Survey Paper Award”, the Best ComSoc Journal Paper
Award in 2021 as well five Best Paper Awards from ICC
and Globecom Conferences. He is the author of ten books Jianyong Zhang received B.S. and Ph.D. degrees in optics
and more than 800 publications. He is also the recipient of from Beijing Institute of Technology, Beijing, China, in 2000
the 2017 IEEE Communications Society Wireless Technical and 2004, respectively. In 2006, he joined the Institute of
Committee (WTC) Recognition Award, the 2018 AdHoc Lightwave Technology, Beijing Jiaotong University, Beijing,
Technical Committee Recognition Award, and the 2019 China, and he is currently an Associate Professor and Doc-
IEEE Communications and Information Security Technical toral Supervisor of the Institute of Lightwave Technology.
Recognition (CISTC) Award. He served as the Editor-in- His research interests include classical and modern coding
Chief of IEEE Network and is currently serving on the theory and information theory as well as their applications
Editorial Boards of many IEEE Transactions and Magazines. in optical and wireless communications. He has authored or
He was the Chair of the IEEE Communications Society coauthored more than 40 academic papers in peer-reviewed
Wireless Technical Committee and the Chair of the TAOS international journals and conferences.
Technical Committee. He served as the IEEE Computer
Society Distinguished Speaker and is currently the IEEE
ComSoc Distinguished Lecturer.

12