Energy-Efficient Service Function Chain Provisioning in
Multi-Domain Networks
Gang Sun
1, 2
, Yayu Li
1
, Guangyang Zhu
1
, Dan Liao
1
and Victor Chang
3
1
Key Lab of Optical Fiber Sensing and Communications (Ministry of Education),
University of Electronic Science and Technology of China, Chengdu, China
2
Center for Cyber Security, University of Electronic Science and Technology of China, Chengdu, China
3
Xi’an Jiaotong-Liverpool University, Suzhou, China
Keywords: Service Function Chain, Energy Efficiency, Provisioning, Multi-Domain Networks.
Abstract: Service Function Chain (SFC) is not only helpful for saving the capital expenditure (CAPEX) and
operational expenditure (OPEX) of network provider, but also can reduce energy consumption in the
substrate network. However, to best of our knowledge, few researches focus on the problem of energy
consumption for provisioning SFC requests in multi-domain networks. In this paper, we firstly formulate
the problem of energy-efficient online SFC request provisioning across multiple domains by using integer
linear programming (ILP). Then we propose a heuristic algorithm called EE-SFCO-MD for efficiently
solving this problem. We conduct simulation experiments for evaluating the performance of our algorithm.
The simulation results show that EE-SFCO-MD performs better than existing approaches.
1 INTRODUCTION
In traditional networks, network functions (e.g.,
Firewall, Network Address Translation, etc.) are
implemented by middle-boxes coupled to the
hardware (Kuo T W et al., 2016). However, with the
development of Internet, the cost for maintaining
these hardware-implemented network functions
becomes higher, and the hardware devices can’t
easily meet the customization demand of end-users.
To mitigate this problem, network function
virtualization (NFV) is proposed to implement
network function on commodity servers, which is
called virtualization network function (VNF) (Pham
C et al., 2017).
Usually, a service function chain (SFC) (Eramo
V et al., 2017) (Elias J et al., 2017) is composed by
several VNFs with specific order. The mapping of
an SFC request is to find several physical servers to
host the VNFs and physical paths to connect the
servers while satisfying various constraints.
Therefore, a good mapping strategy can not only
save CAPEX and OPEX, but also reduce energy
consumption in substrate network. In single domain
network, the mapping-decision maker can obtain the
global information of physical network, which
makes it possible for completing the mapping policy
based on a global perspective. However, in multi-
domain networks, the detail information of each
domain should be confident for other domains or a
third part to keep the privacy of each domain, which
makes it more difficult to provision the SFC request.
There are two ways to address the problem of
SFC provisioning across multiple domains, i.e., the
centralized and distributed approaches. The
centralized approach needs each domain to share its
own information with other domains or a third part,
just like the research (Dietrich D et al., 2017), which
violates the privacy requirements among various
domains. On the other hand, as shown in the paper
(Abujoda A, Papadimitriou P. 2016), the distributed
method keeps the privacy of each domain during the
process of provisioning SFC requests, which also
results in lower performance and longer response
time. Moreover, the authors don’t take energy
efficiency into account.
To reduce energy consumption, the provisioning
result of an SFC should turn on as few servers as
possible (Melo M et al., 2015) (Sun G et al., 2015).
For offline scenarios, the SFC requests are given in
advance and the mapping strategy usually redesigns
the topology of SFC requests to reduce the number
of VNFs by consolidating the same VNFs in
different requests, such as the research in Yang K et
al., 2016. However, for online SFC requests, the
144
Sun, G., Li, Y., Zhu, G., Liao, D. and Chang, V.
Energy-Efficient Service Function Chain Provisioning in Multi-Domain Networks.
DOI: 10.5220/0006770301440152
In Proceedings of the 3rd International Conference on Internet of Things, Big Data and Security (IoTBDS 2018), pages 144-152
ISBN: 978-989-758-296-7
Copyright
c
2019 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
request arrives and leaves dynamically, which makes
it impossible to know all the SFC requests in
advance. Therefore, it is difficult to provision the
online SFC requests just like offline requests.
In this paper, we analyse the power consumption
for provisioning SFC requests in substrate network
and formulate the problem of energy-efficient
provisioning of online SFC requests across multiple
domains by using integer linear programming (ILP).
The model’s objective is to minimize the power
consumption and we also give the constraints which
must be met during mapping an SFC such as
resource constraints and VNF order constraints.
Since the problem of provisioning SFC requests
across multiple domains is NP-hard (Wang Y et al.,
2017), we also propose a heuristic algorithm named
EE-SFCO-MD to efficiently solve the problem. The
EE-SFCO-MD algorithm firstly extends node
aggregation (Hong S et al., 2014) to build a domain-
level function graph of the physical network. Then
the algorithm generates all domain-level reachable
paths between the source and destination of SFC
request. For each domain-level reachable path, EE-
SFCO-MD will build a local candidate graph based
on the path and select a candidate provisioning
solution with minimum energy consumption as the
online SFC request partitioning result on the
domain-level reachable path. And then the bidding
mechanism is used for select the minimal energy
consumption segment solution among various
domain-level reachable paths as the final SFC
partitioning scheme. Finally, EE-SFCO-MD maps
each sub-SFC of the final partitioning result into
corresponding domain.
The remainder of this paper is organizes as
follows. Section 2 describes the problem of energy-
efficiently provisioning for online SFC requests
across multiple domains and formulates the problem
as an ILP optimization problem. In Section 3, we
propose a heuristic algorithm for efficiently mapping
online SFC request. Section 4 gives the simulation
result and analysis. And we conclude this paper in
Section 5.
2 PROBLEM STATEMENT AND
FORMULATION
We research the problem of how to reduce energy
consumption for provisioning an online SFC request
in multiple domains network, which is different
from the single domain network is that the VNFs of
SFC request may be deployed to several different
domains and the virtual link need to be embedded on
an inter-domain physical path if the two endpoints of
SFC are hosted in two different domains.
Unlike offline SFC request, the online SFC
requests arrive dynamically and uncertainly and thus
cannot be accurately predicted, and the number of
SFC requests need to be processed cannot be known
in advance. Therefore, the energy-efficient online
SFC request provisioning problem is not suitable for
the purpose of saving energy by merging VNFs on
the SFC requests as an offline problem.
To save energy, it’s best for the mapping scheme
to share the demand-meet active physical server as
much as possible for activating fewer servers.
However, in multi-domain environment, the biggest
challenge of provisioning an online SFC request is
the provisioning decision makers do not have all the
information in each domain of substrate network. In
addition, the correct order of VNFs in the SFC
request should be kept in the provisioning result.
Moreover, the energy cost (e.g. server computation
power cost and network transmission power cost)
and the response time of requests are expected to be
reduced while all of the resource and function
constraints are satisfied.
2.1 Primary Definitions
2.1.1 SFC Request
We model an SFC request as a directed weighted
graph
=(
,
,,), where
represents
the set of VNFs and
denotes the set of VNFs-
connected virtual links on an online SFC request.
and denote the source and destination of the
SFC request, respectively. For each VNF node
, (
) indicates its computing resource
demand, and (
) denotes the function demand
of
. For each virtual link
, (
)
represents the amount of link bandwidth resource
demanded. We consider a scenario in which each
virtual link on an SFC request has the same
bandwidth resource demand but the computing
resource demand and function demand of each VNF
node are different.
2.1.2 Physical Network
The physical network can be modelled as an
undirected weighted graph
=(
,
), where
indicates the set of physical noes and
represents
the set of substrate edges in physical network. For
each physical node
, the available computing
resource on
can be denoted by (
), and (
)
Energy-Efficient Service Function Chain Provisioning in Multi-Domain Networks
145
represents the function category deployed on node
. What needs to specify is that
(
)
is equal to 0
when the server does not host a VNF. For each
physical edge
, (
) represents its available
bandwidth resource. In multiple-domain networks,
the physical network consists of m domains
connected by several cross-domain edges. We use
=(
,
)(1 ≤ ≤ ) to denote the substrate
network in the i-th domain.
represents the set of
physical nodes in the i-th domain. And the set of
physical edge in the i-th domain is indicated by
.
Moreover,
is used for representing the set of
inter-domain links in
. Therefore, we can also use
=
∪
∪…∪
∪
to denote a multi-
domain networks.
2.2 Extended Node Aggregation
In multi-domain networks, node aggregation
technology facilitates simplifying the substrate
network can clearly show the connectivity of the
domains in the underlying network, which are
critical to orchestrate SFC requests across multiple
domains. However, if we need to take the function
constraints of SFC requests into account, only the
connection between the various domains is not
enough. Therefore, we extend node aggregation to
abstract physical network into domain-level function
graph which is shown in Figure 1(b) for guiding the
SFC request provisioning process in multi-domain
networks.
Figure 1: Extended Node Aggregation. (a)Substrate
network. (b)Domain-Level function graph.
For keeping the privacy and confidentiality of
each domain, only the shared public information is
utilized in the extended node aggregation (ENA)
approach to construct an abstracted network. In
Figure 1(b), the nodes represent the domains in the
substrate network and the solid lines indicate that the
endpoints are connected by at least one inter-domain
physical edge, which both are shared information.
Moreover, the Src and Dst denote the source and
destination nodes which are specified by SFC
request. The numbers in each dotted circle which
connects with each node represent the set of
deployed function types in the corresponding
domain currently. The numbers that are provided by
the domain orchestrator in each domain indicate
which function types have been deployed in current
domain. And the numbers also represent that the
domain can try to deploy the VNFs which have the
same function demand to share servers for the sake
of reducing the number of active servers. Although
the function types in each domain are shared by the
main orchestrator, the specific quantity and location
of the functions in a domain are still confidential to
other domains. Therefore, this ENA process is not
contrary to the privacy of each domain.
In summary, all of the needed information in our
algorithm includes: i) the domains and the
connectivity between domains; ii) the source and
destination node of SFC request; iii) the set of
deployed function types in each domain. Obviously,
our algorithm does not use the specific information
in each domain.
2.3 Energy-Efficient SFC Provisioning
Across Multiple Domains
2.3.1 Energy Consumption
The energy consumption for the online SFC request
provisioning mainly consist of two parts: the
computing energy consumption and the forwarding
energy consumption. The computing energy
consumption is generated by the servers which host
the VNFs. And the forwarding energy consumption
is generated by forwarding the traffic in the network.
If a server is active, it will consume some basic
energy even if it does not host any VNF (i.e. the
server has no workload), which is called basic power
consumption. As the workload increases, the server
will need to consume additional energy to process
the workload. Thus the power consumption of a
server also consists of two parts: the basic power
consumption and the workload-dependent power
consumption. And the energy cost of a physical node
IoTBDS 2018 - 3rd International Conference on Internet of Things, Big Data and Security
146
can be calculated according to the following
formula.
,;
0, .
basic load
server
nn
n
P
P u if server n is on
P
if server n is off
+ ×
=
(1)
Where
represents the power consumption
generated by the server n to process the traffics of
SFC requests;
indicates the power
consumption of a zero-workload server;
denotes the maximum additional energy cost of a
full-workload server; and u is the server workload
(i.e., computing resource utilization). And
is
equal to 0 when the server n is off.
In addition, when a VNF is hosted on a server,
the server needs to communicate with other servers
and forward traffic data to other server, which
consumes some energy. The power consumption of
a server in the forwarding process is related to the
used ports on the server, which can be calculated by
the Formula (2):
,;
0, .
rack port port
link
nnn
n
P
P c if server n is on
P
if server n is off
+×
=
(2)
Where
represents the power consumption
on the server n to forward data. The basic power cost
on a server for data forwarding can be denoted by
; and
indicates the power consumption
for using a port on a server.
represents the
number of used port on server n. What need to
illustrate is that each server needs to use two ports
for hosting a VNF, namely the ports for receiving
and sending data, respectively.
In this paper, the total power consumption of all
the servers to host the VNFs on a SFC request can
be computed as in Formula (3).
R
n server
compute v n
vN
PmP
∈
=×
(3)
Where
is the total computing power
consumption of all the servers to deploy VNFs on a
SFC request;
is a 0-1 variable that indicates
whether the VNF v is deployed on the server n.
=1 if v is deployed on n, and 0 otherwise.
On the other hand, the power consumption for
forwarding traffic of a SFC request between VNF
deployment positions can calculated according to the
following formula.
[
()()]
RS
nk n link k link n
forward ij i n j n i
ij L nk E
server link k server link
nnikk
PmmPmPw
PPwPP
∈∈
=×+×+×
++× +
(4)
where
represents the forwarding power
consumption in substrate network for provisioning a
SFC request.
is mainly composed by two
parts: one part is generated by the physical servers
that have hosted VNFs due to the use of ports for
exchanging data with other servers, such as the first
and second parts of the above equation; and the
other part is generated by the servers for forwarding
purpose. Forwarding nodes not only need to use
server port for forwarding data, but also need to
consume basic power to active the server, such as
the third and fourth parts in the above equation.
represents a virtual link on SFC. is a physical
edge in substrate network.
is a 0-1 variable that
denotes whether the virtual link is embedded on
physical edge . Thus
=1 if is embedded
on , and 0 otherwise.
also is a 0-1 variable
which means whether physical node n is a
forwarding node.
=1 if n is a forwarding node,
and 0 otherwise.
2.3.2 Objective Function
We formulate the online SFC request provisioning
problem by using integer linear programming (ILP)
to minimize the total power consumption. The
objective function is presented in Formula (5).
{
}
compute forward
Minimize P P+
(5)
Objective function tries to minimize the total
power consumption for provisioning an online SFC
request, i.e., the sum of the computing power
consumption and the forwarding power
consumption.
2.3.3 Constraints
The provisioning process of an SFC request must
satisfy many constraints, such as resource capacity
and function constraints, especially in multi-domain
network.
VNF Provisioning:
1,
S
n
vR
vN
mnN
∈
=∀∈
(6)
Energy-Efficient Service Function Chain Provisioning in Multi-Domain Networks
147
1,
R
n
vS
vN
mnN
∈
≤∀∈
(7)
() () () (), ,
R
S
f
nfunv fnfn vN NN×=×∀∈∀∈
(8)
() (), ,
n
vRS
dem v m c n v N n N≤× ∀∈ ∀∈
(9)
Constraint (6) guarantees that a VNF node can
only be deployed on a physical node. Constraint (7)
ensures that the number of VNFs which are hosted
onto a server is less than one in the provisioning
process of the same SFC request. In this paper, we
assume that the number of function types hosted on
a server is no more than one. Thus Constraint (8)
ensures that a server can only deploy the VNFs with
same function. We have to note that
(
)
=0 if the
server n has not deployed any VNF. Therefore, a
VNF can be hosted on the server only if it does not
deploy VNF or its function constraints are satisfied.
Equation (9) is the node resource capacity constraint
of each physical node, which ensures that the server
must have sufficient available resource capacity for
meeting the VNF’s resource demand if the VNF
tend to be hosted on the server.
Virtual Link Provisioning:
() ( ), ,
nk
ij R S
dem ij m b nk ij L nk E≤× ∀∈∀∈
(10)
2,
RS R S
np pn
ij ij S
ij L np E ij L pn E
mmnN
∈∈ ∈∈
+≤∀∈
(11)
Constraint (10) is the bandwidth resource
capacity constraint of each physical edge, which
guarantees that a virtual link’s bandwidth resource
demand must not exceed the available bandwidth
resource of physical edges which host the virtual
link. Due to an SFC request is an ordered chain of
VNFs, the provisioning solution of SFC should be
loop-free to avoid the emergence of Ping-Pong
traffic. Equation (11) restricts a physical node to be
used at most once, which is helpful to obtain acyclic
mapping solutions for SFC.
Order Constraints: since the VNFs are orderly
connected in an SFC request, their deployment
nodes should also be connected at the specific order.
The order constraints between VNFs can be
expressed as in the following formulas.
int int
,
er er
LL
ph hp h h
ij ij j i R
ph E hp E
mmDDijL
∈∈
−=−∀∈
(12)
,
,
hh
SS
kn nk hn hn
ij ij j j i i
kn E nk E
h
SR
mmDmDm
nN ijL
∈∈
−=×−×
∀∈ ∀∈
(13)
,,
nk n n
ij i i R S
mmwijLnkE=+∀∈∀∈
(14)
Where
is a 0-1 variable which represents
whether VNF j is deployed in domain h. And
=
1
if h hosts j and 0 otherwise. Therefore, Equation
(12) ensures that at least one inter-domain edge can
be found to connect the two domains when two
VNFs are allocated to different domains. Constraint
(13) guarantees that the VNF deployment results in
each domain keeps the correct order. Moreover, if
there are forwarding nodes on the underlying
embedding path of a virtual link, the correct order of
VNFs should also be guaranteed, which is
constrained in Constraint (14). Furthermore, the
endpoint of the provisioning solution must be the
destination of the SFC request rather than a
forwarding node, so we just need to constrain the
first node of the embedded edges of each virtual
link. If
=0, server n can neither host any VNF
nor be a forwarding node. And if
=1, the
server n is either a forwarding node or a VNF
deployment node.
3 ALGORITHM DESIGN
Compared with single domain networks, the process
of provisioning an SFC request can be divided into
two key parts in multiple domains network, i.e., the
partition of SFC request for each domain in the
network and the mapping of sub-SFCs in each
domain.
In this section, we propose a heuristic algorithm
called EE-SFCO-MD for energy-efficient
provisioning online SFC requests across multiple
domains where the requests arrive and leave
dynamically. Without loss of generality, we assume
that the online SFC requests arrive and leave
according to a Poisson process in this work. For
each online SFC request, EE-SFCO-MD firstly
extends node aggregation to build a domain-level
function graph (DLFG) based on the substrate
network to keep the privacy of each domain, as is
IoTBDS 2018 - 3rd International Conference on Internet of Things, Big Data and Security
148
shown in Figure 1. Then EE-SFCO-MD generates
all domain-level reachable paths (DLRP) between
source and destination in the domain-level function
graph. As is shown in Figure 1(b), a domain-level
reachable path may go through several domains.
Since the VNFs of an SFC request has the order
constraints, which means the deployment solution
must keep the correct order of VNFs on the SFC.
Moreover, for each domain-level reachable path,
EE-SFCO-MD sends the SFC request to each
domain on the path and collects all the candidate
information from each domain to construct a local
candidate graph (LCG). Based on the local candidate
graph, EE-SFCO-MD partitions the SFC request into
several sub-SFCs with minimum energy
consumption for the domains in domain-level
reachable path. And then the bidding mechanism is
used for obtaining the final SFC request partitioning
result among various domain-level reachable paths.
Finally, EE-SFCO-MD maps the sub-SFCs of the
final partitioning result into corresponding domains.
The pseudo code of the EE-SFCO-MD algorithm is
shown in
Algorithm 1.
Algorithm 1: Energy Efficient SFC
request Orchestrating across
Multiple Domains (EE-SFCO-MD).
Input:
(1) Substrate network
=
(
,
)
;
(2) Online SFC request
queue,ArrivedSFC;
Output:
The set of accepted SFC requests SFC
acc
and the mapping solutions M
SFC
.
1: Initialization:
= ∅,
=∅;
2: while ≠ ∅,do
3: Query SFC requests in DeployedSFC
and release the resources occupied
by the expired SFC requests, and
then remove the expired SFC
requests from DeployedSFC;
4: Take out the first SFC request
sfc
1
in ArrivedSFC and construct
the domain-level function graph
according to sfc
1
, and then
generate all domain-level
reachable paths PATH between
source and destination of sfc
1
,
let the optimal mapping
=∅;
5: for each ℎ ∈ ,do
6: Construct local candidate graph
lcg based onℎ, and generate the
partitioning result of sfc
1
according to lcg;
7: Based on the partitioning result,
premap the sub-SFC in each domain
onℎ;
8: if find an provisioning solution M
successfully, do
9: if the energy consumption of M <
, do
10:
←;
11: endif
12: endif
13: endfor
14: if
≠∅, do
15: Map sfc
1
into substrate network
according to
, and update the
substrate network;
16:
=
∪
,
=
∪
,
= ∪
;
17: endif
18: ArrivedSFC = ArrivedSFC \{sfc
1
};
19: endwhile
20: return
,
.
In Algorithm 1, all arrived online SFC requests
are firstly buffered in a queue named ArrivedSFC.
The notation DeployedSFC is used for representing
the set of deployed online SFC requests. When
ArrivedSFC ≠ ∅ , each online SFC request in the
ArrivedSFC queue is deployed one by one (line 2).
Due to the limitation of physical resource, the online
SFC requests may be blocked. Therefore, we define
to denote the set of accepted SFC rquests.
Before deploying a new online SFC request, our
algorithm firstly query the expired request in
DeployedSFC and remove them from DeployedSFC,
and release the physical resource occupied by these
expired SFC requests (line 3). And then, EE-SFCO-
MD constructs the domain-level function graph
according to the first SFC request in the ArrivedSFC
and generates all domain-level reachalbe paths (line
4). Line 5-13 in
Algorithm 1 are responsible for
obtaining the final SFC partitioning solution with
minimum energy consumption. Due to the space
limitation, we omit the algorithms for constructing
local candidate graph and candidate selection (line
6). Moreover, EE-SFCO-MD deploys each sub-SFC
in the final segment result into substrate network
(line 14-15). Finally, EE-SFCO-MD updates the
substrate network and the sets
,
,
and ArrivedSFC (line 16-18).
Energy-Efficient Service Function Chain Provisioning in Multi-Domain Networks
149
4 SIMULATION RESULT
In this section, we numerically compare the
performance of EE-SFCO-MD algorithm with the
algorithm proposed in (Abujoda A, Papadimitriou P.
2016) and (Wang Y et al., 2017). We first describe
the simulation environment, and then present our
simulation results and analysis.
4.1 Simulation Environment and
Settings
We generate the multiple domains physical network
by using IGEN tool. The substrate network is
composed by 6 domains and each domain has 20
nodes. In each domain, the physical nodes are
connected by the Delaunay model in IGEN. And the
domains are connected by inter links with a
probability of 0.5. The resource demands of VNFs
and virtual links both follow a uniform distribution
U (10, 40). The computing resource capacity of
physical nodes and bandwidth resource capacity of
physical intra-domain edges follow a uniform
distribution U (200, 300), and the bandwidth
resource capacity of physical inter-domain edges
follows a uniform distribution U (4000, 6000).
Moreover, a server’s zero-workload power and full-
workload power are set to 171W and 301W,
respectively. The basic power cost for forwarding on
a server is set to 5W, and the power consumption for
using a port on a server is set to 1.2 W.
4.2 Simulation Result and Analysis
The average response time of EE-SFCO-MD is
lower than that of Nestor and DistNSE algorithms,
as shown in Figure 2. This is because our algorithm
extends the node aggregation for constructing a
domain-level function graph of the substrate
network, which is helpful for guiding the
provisioning process. Moreover, EE-SFCO-MD just
needs to search on the intra-domain function graph
whose scale is much smaller than the substrate
network. Therefore, EE-SFCO-MD can quickly
response to the online SFC requests. It needs to be
mentioned that DistNSE algorithms traverses all the
physical paths in each domain, which results in a
significant increase in response time.
23456
0
30
60
90
120
3000
3500
4000
4500
5000
Average Response Time (ms)
Length of SFC
Nestor
DistNSE
EE-SFCO-MD
Figure 2: The average response time as a function of the
length of SFC.
23456
6000
8000
10000
12000
14000
16000
Average Server Energy Consumption (J)
Length of SFC
Nestor
DistNSE
EE-SFCO-MD
Figure 3: The average server energy consumption as a
function of the length of SFC.
As shown in Figure 3, the average server energy
consumption of EE-SFCO-MD algorithm is much
lower than that of Nestor and DistNSE. This is
because that EE-SFCO-MD takes reusing servers
into account, which makes it possible for EE-SFCO-
MD to find the minimal energy consumption servers
for hosting VNFs while provisioning online SFC
requests. Whereas Nestor and DistNSE algorithms
don’t consider to saving energy during mapping SFC
requests, which leads to a high energy cost.
IoTBDS 2018 - 3rd International Conference on Internet of Things, Big Data and Security
150
23456
200
400
600
800
1000
Average Link Energy Consumption (J)
Length of SFC
Nestor
DistNSE
EE-SFCO-MD
Figure 4: The average link energy consumption as a
function of the length of SFC.
23456
6000
8000
10000
12000
14000
16000
18000
Average Total Energy Consumption (J)
Length of SFC
Nestor
DistNSE
EE-SFCO-MD
Figure 5: The average total energy consumption as a
function of the length of SFC.
Similarly, by searching on the intra-domain
function graph, EE-SFCO-MD can find the shorter
physical paths for embedding the virtual links on the
SFC request, which can reduce the energy
consumption for forwarding data between servers,
which is shown in Figure 4.
Figure 5 describes the average total energy
consumption among three compared algorithms.
From Figure 5, we can see that the EE-SFCO-MD
algorithm has lower total energy consumption. This
is because that EE-SFCO-MD algorithm considers
saving energy during provisioning online SFC
request.
5 CONCLUSIONS
In this paper we study the problem of energy-
efficient provisioning for online SFC request in
multi-domain networks. We firstly formulate the
problem as an optimization problem by using ILP
and propose a heuristic algorithm named EE-SFCO-
MD for solving this problem. The simulation results
show that our algorithm is promising for reducing
energy consumption and perform better than existing
approaches.
ACKNOWLEDGEMENT
This research was partially supported by the National
Natural Science Foundation of China (61571098),
Fundamental Research Funds for the Central Universities
(ZYGX2016J217).
REFERENCES
Kuo T W et al., 2016. Deploying chains of virtual network
functions: On the relation between link and server
usage. In IEEE INFOCOM 2016 - the IEEE
International Conference on Computer
Communications. 1-9.
Pham C et al., 2017. Traffic-aware and Energy-efficient
vNF Placement for Service Chaining: Joint Sampling
and Matching Approach. In IEEE Transactions on
Services Computing.
Eramo V et al., 2017. An Approach for Service Function
Chain Routing and Virtual Function Network Instance
Migration in Network Function Virtualization
Architectures. In IEEE/ACM Transactions on
Networking, 99:2008-2025.
Elias J et al., 2017. Efficient Orchestration Mechanisms
for Congestion Mitigation in NFV: Models and
Algorithms. In IEEE Transactions on Services
Computing, 10(4):534-546.
Dietrich D et al., 2017. Multi-Provider Service Chain
Embedding With Nestor. In IEEE Transactions on
Network & Service Management, 14(1):91-105.
Abujoda A, Papadimitriou P. 2016. DistNSE: Distributed
network service embedding across multiple providers.
In International Conference on Communication
Systems and Networks. IEEE, 1-8.
Wang Y et al., 2017. Cost-Efficient Virtual Network
Function Graph (vNFG) Provisioning in Multi-Domain
Elastic Optical Networks. In Journal of Lightwave
Technology, 99:1-1.
Melo M et al., 2015. Optimal virtual network embedding:
Energy aware formulation. In Computer Networks,
91(C):184-195.
Energy-Efficient Service Function Chain Provisioning in Multi-Domain Networks
151
Sun G et al., 2015. Power-Efficient Provisioning for
Online Virtual Network Requests in Cloud-Based Data
Centers. In IEEE Systems Journal, 9(2):427-441.
Yang K et al., 2016. Energy-Aware Service Function
Placement for Service Function Chaining in Data
Centers. In Global Communications Conference
(GLOBECOM). IEEE, 1-6.
Hong S et al., 2014. Virtual optical network embedding in
multi-domain optical networks. In Global
Communications Conference (GLOBECOM). IEEE:
2042-2047.
IoTBDS 2018 - 3rd International Conference on Internet of Things, Big Data and Security
152
