A Method of Service Composition in Software-

Defined Networking

Olena Tkachova Ievgen Duravkin

Ph.D., Infocommunication engineering department D.Sc, Infocommunication engineering department
Kharkov National University of Radio Electronics Kharkov National University of Radio Electronics
Kharkov, Ukraine Kharkov, Ukraine
[email protected] [email protected]

Hassan Mohamed Muhi-Aldeen

Postgraduate student, Telecommunication system department
Odessa National Polytechnic University
Odessa, Ukraine
[email protected]

Abstract — The work is devoted to developing the One of the main tasks of the control system is supporting
complex methods of service composition and provision in the availability and reliability of services. The solution of the
Software-Defined Networking. The service provision task usually based on structural network redundancy and
mechanism in the multiservice network under Software- resources reservation. Such approach leads to an unjustified
Defined Networking concept is analyzed. An overview of increasing the cost of provided services without supporting
existing service provision methods that are widely used in necessary level of QoS. The developing the methods of
distributed computing networks is proposed. The solution complex services provision with resources optimization that
of the multicriteria task of services selection and allows supporting required QoS characteristics is a relevant
task.
composition with optimal resources allocation are
proposed in the work. The model of complex service The analyses of service composition mechanisms and
provision in Software-Defined Networking and algorithm resource distribution effectivity and methods of increasing the
of service composition with optimal resource allocation are quality of complex services in distributed network are
also represented. proposed in work [6-10]. The work [6] is devoted to solving
the problem of search services and the allocation of resources
Keywords—complex service; Software-Defined Networking; in order to minimize the total time of service provision. In [6,
QoS criterion; resources allocation; service provision 7] introduced a new criterion for assessing the quality of
services - processing requests delayed consideration, arising
I. INTRODUCTION from the geographical remoteness of the resource from the
consumer. The method based on genetic algorithm to choose
The Software-Defined Networking (SDN) is one of the the resources taking into account the geographical distribution.
more perspective technology of organization and management In [8] the resources allocation is based on QoS characteristics
of modern info-communication networks. The main idea of such as the response time, cost, availability, and reliability.
SDN is separation the management functions from data The task is formulated as an optimization problem with
transfer functions [1-3]. In accordance with the SDN concept, constraints on the QoS metrics. Various schemes such as
all management and monitoring functions are transferred to a sequential execution, cyclical, parallel and selective of service
separate central unit - the controller [3, 4]. The controller composition are considered in [9, 10]. The resulted service
generates an optimal solution for services provision based on selection is based on such metrics as reliability and
information about the network topology and current availability.
characteristics. Controller together with management modules
forms the control system. SDN network infrastructure includes The above approaches are widely used in the management
both physical and virtual elements. This allows significantly of a distributed network infrastructure. However, their direct
improve the performance and scalability of the network. Rapid application in the SDN is difficult for several reasons related
deployment, flexible combination of multiple network to the high cost of network resources in the allocation process,
functions on a single servers’ platform, automatization of lack of registration of computing elements, a limited number
administrative procedures are the main advantages of SDN of analyzed QoS metrics and possible solutions of resources
[5]. allocation. Thus, the solution of the multicriteria task of
services selection and composition with optimal resources
allocation is relevant. An effective solution to the task allows
increasing availability and ensures the required quality of (4) Controller chooses the optimal route (OpenFlow
service in SDN. The model of complex services provision Switch-to-OpenFlow Switch) and creates a VLAN after the
with a flexible choice of QoS criterions in SDN, and method resource for service provision is founded.
of optimal resources allocation based on QoS requirements are
proposed in the work. (5) Controller generates reply message with information
required to access the service – computation resources
II. THE MODEL OF SERVICES PROVISION IN SOFTWARE- allocation (IP address of computation nodes).
DEFINED NETWORKING
The structural scheme of SDN architecture and allocation
Multiservice networks, constructed under the SDN of resources for service provision process are shown on Fig.1.
concept, often have a complex structure; computation and
forwarding elements such network are geographically
distributed. The controller executes the gathering and analyses
of information about current network topology and network
characteristics on the control plane. OpenFlow switches via
OpenFlow Protocol [11, 12] provide the information about
active network components. The controller has no complete
map of the network infrastructure; the network map includes
an only limited set of logical networks called VLANs (Virtual
Local Area Network, VLAN). VLANs that are created
dynamically and forming abstractions as overlay networks [3].
In such case the process of resources allocation for service
provision includes the next step [1, 2, 13]:

(1) End user sends a request for service provision.

Regarding Service Layer Agreement (SLA) each service has
his own requirements to quality of service provision: response
time, throughput, reliability, etc. The requirements are taken
into account in Type of Services (ToS) fields.

(2) Requests from end users with different intensity

arrive at the ports of OpenFlow switches. OpenFlow switch
parses the request, analyses ToS, Source, and Destination
fields in the header of the request and forwards the request to Fig. 1. Model of resource allocation process for SDN architecture
the Proxy/Web- server or computation node directly if IP
address of those devices is matched as a destination. In case if The proposed fragment of SDN (Fig. 1) is divided into tree
any field was not mashed OpenFlow controller forward LANs: LAN1, LAN2 and LAN3. Each LAN includes a tree
request to the controller. logical groups of network nodes – group of end users’ nodes,
group of compute nodes combined the mix of physical (PM)
(3) Controller performs a service orchestration regarding and virtual computation resources (VM), group of servers
end user requirements. If service is complex and includes a set (Web, Mail, FTP, etc.).
of atomic services the task scheduler analyses QoS The information about services that can be delivered by the
requirements for each atomic servers, put the right order of nodes stores in Service Registry (SR). In this case, each LAN
atomic services and find the proper resource for service has its own SR. The global information about network, in
provision - the controller analyses the current network particular, the information about services that provided by
resources, determines what compute node can satisfy the different LANs stored in Global Service Registry (GSR) and
users’ requirements. placement on Control Layer [9].
(3a) Find the logical network where the services are Assume that a set of end users is defined as
allocated. In the case of using the virtual resources, such U = { u1 ,u2 ,...,um } , a set of services that can be provided in
characteristics as transparency interaction and interoperability SDN defined as TS = {TS1 ,TS2 ,...,TSп } . Amount of services
between VM, good compatibility with hardware that impact to and their QoS is predetermined by SLA [10].
further availability of service. When the appropriate software The information about the services stored in local Data
is found the compatible hardware resources selected. Base in SR and in global Data Base in GSR. Amount of
compute nodes within LAN is defined as N = { n1 ,n2 ,...,nm }
(3b) The hardware resources can be allocated both in the
same LAN with the user or remotely LAN. The choice of and includes a set of virtual machines NVM and physical
LAN for service provision based on the requirements for machines N PM , N  { N PM ,NVM } .
quality of provided services and information about current The information about physical and virtual nodes is saved
characteristics of computation nodes. in local Data Base on Data Layer and the information about
virtual machine is centralized and stored in global Data Base
on Control Layer.
 The policy regulated the service provision process based The global information about services stored in GSR
on chosen QoS metrics. In accordance with IETF GSR[TS ] can be represented by follow:
recommendation [26] for service provision, the next metrics
are identified: cost efficient, performance utilization, response GSR[TS ] = { SR1 [TS ],SR2 [TS ],...,SRq [TS ]...,Rn [TS ]}
time, delay variation, reliability. In the case of using SDN . (4)
architecture information about each service stored in service SR[TS ],SRi [TS ] = {TS1 ,TS2 ,...TSi }
registry characterized by follows:
The model of SDN architecture where is taking into
SR j [TSi ] = { loc(TSi ),d(TSi ),c(TSi ),t(TSi ), p(TSi ),r(TSi )}; account the location of the service registry is shown in Fig. 2.
The set of services that are available in LAN1 is stored in
tki ,loc(TSi ) = U loc( tski ); SR1[TS], LAN2 – SR2[TS], LAN3 – SR3[TS], global service
k
registry is matched as GRS[TS].
d(TSi ) = max d( tski );
k
p
(1)
c(TSi ) =  c( tski );
k =1

p(TSi ) = min p( tski );

k

r(TSi ) = min r( tski );

k

where SR j is a registry where stored information about

service; TS i is complex services that include combination of
monatomic TSi = { ts1i ,ts2i ,...,tski } , loc( TS i ) is IP address
of hardware resource that provided TS i service; с( TSi ) is a
service provision cost efficiency; t( TSi ) is average response
time of service provision; p( TSi ) is a performance of
compute node required to the service execution; d( TSi ) is a
delay variation during service provision. Two factors affect to
the value of delay variation – computation delay and
transmission delay; r( TSi ) is a reliability of service
provision; tski is a monatomic service.
The equation (1) shows the characteristics in the case of
atomic services static placement of computing resources. In Fig.2. Model of service orchestration process
the case of virtual computing resources change dynamically,
TS i service may include a set of monatomic services. To
then compute node performance in the provision of services is
calculated based on the following statements: provide the services with guarantee quality the next
conductions should be taking into account in resources
k
allocation process:
p(TSi ) = p( N PM ) − 
j =1, j  i
p(TS j ) , (2)
III. SERVICE PROVISION METHOD WITH OPTIMAL
where p( N PM ) is the nominal value of the compute node RESOURCE ALLOCATION
performance; p(TS р ) is a compute node performance that are The set of complex service realizations by the composition
of monatomic services is a finite [14]. The QoS resulted
required for р-th complex service.
complex service should corresponding SLA requirement:
The complex service characterized by a set of metrics or n

criterions. The integral function is used for estimate the value FQoS (TSi ) = v1  ts1i + v2  ts2i + ... + vk  tski =  vk  tski → max;
k =1 (5)
of each metrics. The integral function represents a weighted
FQoS (TSi )  FSLA (TSi ).
sum of criterion multiplied by corresponding coefficients:
IF( TSi ) = K1 d( TSi ) + K 2 c( TSi ) + K 3t( TSi ) + n

+ K4 p( TSi ) + K5 r( TSi ),
(3) Fc (TSi ) = c1  ts1i + c2  ts2i + ... + ck  tski =  ck  tski → min;
k =1 (6)
where K1 ,K 2 ,...,K 5 is value of each criterion. The decision Fc (TSi )  Fnom (TSi ),
about service composition is based on value of each criterion
and end user requirements. where FQoS (TSi ) is resulted quality of complex service;
Each SR has its own responsibility of services provision FSLA ( TSi ) is quality of provided services corresponded SLA
that determined by the network service provider and can be requirements;
changed by devices configuration and management module.
v1 ,v2 ,...,vk c ,c ,...,ck However, along with the many benefits remain a number
is a value of each monatomic service, 1 2
of unresolved problems. One of these tasks is to ensure the
is a cost of each monatomic service; Fc ( TSi ) is current
required level of service availability, the effective solution that
meaning of complex service provision; Fnom ( TSi ) is the is the use of complex mechanisms of services and resource
nominal cost corresponded SLA requirements. allocation.
In final decision about resource allocation, the controller The proposed method of service provision with optimal
selects the set of services with minimal cost and maximal resources allocation is the solution of the multicriteria task,
QoS. The monatomic services in (2) can be replaced by other where the take into account a contribution of each monatomic
monatomic services with value that is more appropriate. services in complex service. Application of the algorithm
The flow-chart of proposed algorithm of complex service allows providing the required QoS of the requested service
provision with optimal resource allocation is depicted in Fig.3. due to optimal resource allocation both regarded local service
registry or remote service registry.
REFERENCES

[1] F. Akyildiz, “A roadmap for traffic engineering in SDN-OpenFlow

networks”. Computer Networks, 2014, Vol. 71., pp. 1-30.
[2] Architecture SDN. Open Networking Foundation, 2014. Available at:
https://www.opennetworking.org/
[3] T. Egawa, “SDN standardization Landscape from ITU-T Study Group”.
ITU Workshop on SDN. 2013, Geneva, Switzerland, 4 June, 22 р.
[4] Qinghong Zhong, Ying Wang, Wenjing Li, Xuesong Qiu, "A min-cover
based controller placement approach to build reliable control network in
SDN", Network Operations and Management Symposium (NOMS) 2016
IEEE/IFIP, pp. 481-487, 2016, ISSN 2374-9709.
[5] Daphne Tuncer, Marinos Charalambides, Stuart Clayman, George Pavlou,
"Adaptive Resource Management and Control in Software Defined
Networks", Network and Service Management IEEE Transactions on,
vol. 12, pp. 18-33, 2015, ISSN 1932-4537.
[6] Shengcai Liu, Yufan Wei, Ke Tang, A. K. Qin, Xin Yao, "QoS-aware
long-term based service composition in cloud computing", Evolutionary
Computation (CEC) 2015 IEEE Congress on, pp. 3362-3369, 2015.
[7] Qin Shengjun, Chen Yan, Mu Xiangwei. “An Optimal Service Selection
with Constraints Based on QoS.” Physics Procedia. 2012. Vol. 25,no. 0.
P. 2050–2057Available at:
http://www.sciencedirect.com/science/article/pii/S1875389212007651.
[8] S. Sezer, S. Scott-Hayward, and P. Chouhan, “Are we ready for SDN?
Implementation challenges for Software-Defined Networks”.
Communications Magazine, IEEE, vol. 51, no. 7, July 2013. pp. 36–43.
[9] M. E Michael, P. Singh Munindar, “Agent-based trust model involving
multiple qualities”. Proceedings of the fourth international joint
conference on Autonomous agents and multiagent systems,
Fig.3. Proposed algorithm of complex service provision AAMAS99’05, New York, NY, USA, pp. 519–526, 2005

The proposed algorithm service provision with optimal [10] Shengcai Liu, Yufan Wei, Ke Tang, A. K. Qin, Xin Yao, "QoS-aware
long-term based service composition in cloud computing", Evolutionary
resource allocation includes next steps: Computation (CEC) 2015 IEEE Congress on, pp. 3362-3369, 2015.
1. Definition of the monatomic services that include
required complex service. [11] Nebil Ben Mabrouk, Nikolaos Georgantas, Valérie, "Set-Based Bi-level
Optimisation for QoS-Aware Service Composition in Ubiquitous
2. Definition the optimal composition of monatomic Environments", Web Services (ICWS) 2015 IEEE International
services based on value integral function. Conference on, pp. 25-32, 2015.
3. Reducing the set of obtained composition according
[12] R. Paul, S. Network, “Virtualization and Software Defined Networking
to the value of criterion defined in (2) and (3) and local for Cloud Computing: A Survey”. IEEE Commun. Mag. 2013, Vol.51,
optimum of resources allocation definition. pp.24–31.SDN-NFV Reference Architecture v1.0. Verizon Network
Infrastructure Planning.Verizon, 2016, 220 р.
CONCLUSION
[13] OpenFlow Switch Specification. Version 1.3.0 (Wire Protocol 0x04).
Application SDN concept in multiservice distributed Open Networking Foundation, 2012. Available at:
https://www.opennetworking.org/images/stories/downloads/sdn-
networks can significantly improve the quality of end-user resources/onf-specifications/openflow/openflow-spec-v1.3.0.pdfGerardo
services. The improvement is achieved by forming a
[14] Quanwang Wu, Fuyuki Ishikawa, Qingsheng Zhu, Dong-Hoon Shin,
centralized management system and reduces the load on the "QoS-Aware Multigranularity Service Composition: Modeling and
data network by separating the control plane from the data Optimization", Systems Man and Cybernetics: Systems IEEE
plane. Transactions on, vol. 46, pp. 1565-1577, 2016, ISSN 2168-2216.
on, vol. 46, pp. 1565-1577, 2016, ISSN 2168-2216.