micro operator design pattern in 5g sdn/nfv network

Research ArticleMicro Operator Design Pattern in 5G SDNNFV Network

Chia-Wei Tseng 1 Yu-Kai Huang 1 Fan-Hsun Tseng 2 Yao-Tsung Yang 1

Chien-Chang Liu1 and Li-Der Chou 1

1Department of Computer Science and Information Engineering National Central University Taoyoan 32001 Taiwan2Department of TechnologyApplication andHumanResourceDevelopment National TaiwanNormalUniversity Taipei 10610 Taiwan

Correspondence should be addressed to Li-Der Chou cldcsiencuedutw

Received 27 February 2018 Accepted 30 May 2018 Published 10 July 2018

Academic Editor Shao-Yu Lien

Copyright copy 2018 Chia-Wei Tseng et alThis is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

The trend of 5G mobile networks is increasing with the number of users and the transmission rate Many operators are turningto small cell and indoor coverage of telecom network service With the emerging Software Defined Networking and NetworkFunction Virtualization technologies Internet Service Provider is able to deploy their networks more flexibly and dynamicallyIn addition to the change of the wireless mobile network deployment model it also drives the development trend of the MicroOperator (120583O) Telecom operators can provide regional network services through public buildings shopping malls or industrialsites In addition localized network services are provided and bandwidth consumption is reduced The distributed architecture of120583O tackles computing requirements for applications data and services from cloud data center to edge network devices or to themicro data center of 120583OThe service model of 120583O is capable of reducing network latency in response to the low-latency applicationsfor future 5G edge computing environmentThis paper addresses the design pattern of 5Gmicro operator and proposes a DecisionTree Based Flow Redirection (DTBFR) mechanism to redirect the traffic flows to neighbor service nodes The DTBFR mechanismallows different 120583Os to share network resources and speed up the development of edge computing in the future

1 Introduction

Many new operators have chosen to turn to small cells andmicro telecommunication network service that covers indoorareas in the wake of the coming of the 5G era the changesof habits of users when going online and the increasingdemands for applications Telecom operators can provideservices via a variety of access networks such as 3G 4G 5Gand even Wi-Fi [1ndash3] With the emerging Software DefinedNetworking (SDN)Network Function Virtualization (NFV)technologies Internet Service Provider (ISP) is able to deploytheir networks more flexibly and dynamically [4ndash6] ISPsare now capable of providing localized services by meansof public buildings shopping malls or industrial facilitiesand this has not only changed the model of distributingand establishing mobile network but also gave birth tothe idea of the micro operator (120583O)[7ndash9] Factors such asscarcity of spectrum resources the effective employment ofbandwidth the scale of the mobile market and consumersrsquooptions for the diversity of services conspire to produce the

inception of Mobile Virtual Network Operator (MVNO)which in turn brings more diverse applications and servicesto the telecommunication market [10] MVNO can referto any telecommunication operators who provide wirelesscommunication services to consumers and do not have theirown wireless network infrastructures In order to expandthe coverage of their business Mobile Networker Operators(MNO)might choose to cooperate with otherMNOs by rent-ing the bandwidth and time of use in their domains to rapidlydevelop the business of MVNO MVNO provides brandnew opportunities for development for the matured mobilecommunication market and spurs the telecommunicationmarket to move toward a service-oriented competition Asfor 5G mobile network operators development of small cellsand micro telecommunication network service that coversindoor areas is now being initiated to satisfy usersrsquo demandfor indoor coverage maturing the concept of 120583O [11ndash13]

Figure 1 shows the 120583O network architecture 120583O is rela-tively small in scale and holds limited resources to provideparticular and necessary services to a certain number of

HindawiWireless Communications and Mobile ComputingVolume 2018 Article ID 3471610 14 pageshttpsdoiorg10115520183471610

2 Wireless Communications and Mobile Computing

Micro OperatorInternet

3G

LTEMacrocell

Micro DataCenter

Firewall

Firewall

Data Center

VoIPServer

CloudVirtualizedServers

HTTPServer

StreamingServer Application

Server

DSNAAAServer

SDNOpenFlowController

WiFi IEEE80211

abgnp

IP Tunnel

Figure 1 Micro operator network architecture

users The 120583O is originated from Oulu University in Finlandthat consists of three features [14] (1) Mobile devices accesslocal network specific service (2) spatially confined areawith specific service (3) dependent on appropriate availablenetwork resources Facilities like hospitals schools largeconference sports venues shopping malls hypermarketsand factories can use local area network service to satisfyusersrsquo needs for all sorts of application services

Restricted by resources and space 120583O has the characterof regionalization service that allows it to provide undercircumstances where hardware infrastructures are scarceand the resources are limited different regional servicesaccording to different domains making it possible for usersof mobile devices to gain access to services in the nearbyregions Apart from reducing the consumption of bandwidthresources by offering nearby network services this kind oflocalization service can also transfer applications data andthe computing of services from nodes in the data centeron cloud to edge nodes in logical LAN to be processedand realize the developing environment for edge computing[15 16] reduce network latency and satisfy 5G demands foramelioration of latency

The purpose of this paper is to discuss the 120583O designpattern in 5G SDNNFV network and establishes an experi-mental environment to evaluate the decision tree based trafficflow redirectionmechanismTo realize the customized exper-iment environment the network functions virtualization istaken into consideration in our research Our contributionsare listed as follows

(1) A120583Odesign pattern is constructed based on SDNandNFV that combines network slicing and tunnelingtechnologies to build a network infrastructure for120583Os

(2) A DTBFR mechanism for 120583O is proposed and thedecision tree theory is utilized to serve as the referencein the determining the SDN traffic flows direction

(3) We established a 120583O environment for validation ofthe DTBFR mechanism and 120583Os communicationexperiments

The rest of the paper is organized as follows in Section 2the background and related works are addressed Section 3describes the 120583O design pattern and DTBFR mechanismSection 4 presents the experiment results The last sectionconcludes this paper

2 Background and Related Works

The development of 5G application services will be largelyon the Internet of Things (IoT) and encourages the newcommunication market to move toward a more vertical-subdivision one [17 18] The result will be the formation ofdifferent emerging application service scenarios and morediverse demands for networks However as far as telecom-munication is concerned despite the fact that there is nowa globally agreed requirement on IoT constructed by 5G oncharacters such as transmitting speed capacity coverage andsecurity there is still room for the business model of 5Gto improve itself 5G mobile broadband network puts theaccent on small cellsbase stations how to amplify indoorcoverage providing faster services for users and reducingthe latency in network transmission all of which pose agreat challenge for telecommunication operators [19ndash21]Furthermore the appearance of MVNOs has brought newopportunities for development for MNOs who have not yetgained licenses to operate mobile communication business

Wireless Communications and Mobile Computing 3

Wholesale services Retail services

Vertically defined servicesContext-defined services

Micro Operator

Venues Hotels Offices

Service Model Provides local connectivity to MNOsrsquo customers

Service Model Provide local contentand on-demand connectivity to deviceowners or tenants

Service Model Tailor-made on-demandservice for equipment tenants and systems

Service Model Provide users with individual customized content and on-demand services

Venues Personalized consumer services Venues Manufacturing sites smart grids and Hospitals

Venues Shopping malls and Campuses

(O)

Figure 2 Micro operator service model [24]

MVNOs use MNOsrsquo spectrum and network to provide cus-tomized mobile communication services the mobile virtualprivate network for enterprises or scores of other micromarkets that operators still have not stretched their servicesto such as less lucrative or regional local emerging marketsA trend of regionalization has been observed among smallcellbase stations and the regional services do not processesthe characters of public area network industrial applica-tion network and a hybrid network 120583O is an emergingnetwork business model that has started developing underthis backdrop [22] Reference [14] illustrates the relationbetween 120583O and MNO and scrutinizes the use cases of 120583O-MNO to set the foundation for the possible business modelfor 120583O Reference [23] proposes a mechanism in regard to120583Orsquos shared spectrum access communication and shares theinfrastructure of mono-networkphysical network via virtualtechnology to fully utilize the valuable bandwidth resource

As shown in Figure 2 the service model of 120583O can bedivided into four quadrants

(1) The first quadrant is closed controlled independentand general service that often takes places in officeThe users in this region can gain access to the servicesprovided by the region

(2) The second quadrant is shared transparent andgeneral service that often takes places in shoppingmalls and schools Such regions are often packed withpeople and a great deal of data is required as well Amicro data center might be set in the region for usersto quickly gain access to the services

(3) The third quadrant is shared transparent and spe-cial service that often takes place in scenarios liketelematics network and customized services Suchservices are likely to be used by end-users Thecommunication among end-users requires specialnetwork protocols

(4) The fourth quadrant is closed controlled indepen-dent and special service that might take place inhospitals automated factories and smart grids Hos-pitals provide irreplaceable medical services Users

will have to connect to such regions before gainingaccess to the services

Due to the increasing demand of mobile device usersaccessing Internet services operators need to dynamicallyadjust and combine to meet the requirements of differentapplications in order to improve network performance Theservice architecture of 5G 120583Omust also be developed in con-junction with different technologies such as SDN and NFV[25 26] SDN abstracts the network architecture by decou-pling the network control and forwarding functions enablingthe network control to become directly programmable andthe underlying infrastructure to be abstracted for applicationsand network services [27] By separating the control planefrom the data forwarding plane and virtualizing all of the con-nections administrator can remove the hard-wired barriersof networks and quickly change structures to suit their needsIn addition to the SDN NFV is a core structural change inthe way telecommunication infrastructure gets deployedThegoal of NFV is to enable service providers to reduce costs andfaster service delivery The requirements and open standardsthat underpin NFV are being developed by the EuropeanTelecommunications Standards Institute (ETSI) [28]

Since 5G technology is now undergoing rapid develop-ment the conventional network infrastructure is not able tomeet 5Grsquos diverse demands and 120583Osrsquo development of servicemodel in the future Therefore the international standardsorganization and equipment manufacturers now set out topromote the technology of network slicing which allows ISPto use SDN and NFV technologies to realize network virtualslicing the division of several different service scenarios andthe provision of customized network service [29ndash31] The 5Gwhite paper proposed by Next Generation Mobile Networks(NGMN) that expounds 5Grsquos network infrastructure in thefuture had included the concept of network slicing technol-ogy in order to support particular connection forms and useparticular methods to deal with the communication servicesof controluser-plane [32] To satisfy the requirements of5G networkrsquos flexible configuration NOKIA had come upwith the programmable 5G network infrastructure wherenetwork slices of several independent virtual subnetworks


Core network

BG

BGOVS

OVS

OVSOVS

BG

Micro Operator

Datacenter

BGBG

3G

4GLTE

OpenVirtex

tunnel

tunnel tunnel

tunnel

tunnel

NFVI

NFVI

NFVIuO SlicesDC Target Services

Slice 3

Slice 2

Slice 1

SDNOpenFlow Controller


tunnel

Figure 3 A 120583O design pattern with network slicing

are constructed in the same basic infrastructure to servethe purpose of the specific application [33] Network slic-ing technology can divide network resources into multipleindependent virtual networks Not only can it satisfy thediverse demands of small cellsbase stations and 120583O but itcan also work as the crucial technology for the developmentof 5G network communication As for virtualization tools inorder to realize the function of network slicing [34] proposesthe concept of Network Virtualization LayerndashFlowVisor thatdivides a physical network into several virtual networks thateach possesses their own controllers and enables personalvirtual network to be built in the same physical networkequipment to achieve the goal of resource allocation Ref-erence [35] proposes a platform of network virtualizationcalled OpenVirtex (OVX) that is capable of realizing networkvirtualization for multiple tenants OVX can map a userrsquosnetwork topology to a physical topology according to theuserrsquos needs and accomplish the network-connecting processThe major difference between OVX and FlowVisor is thatusers of OVX do not have to worry about the basic physicaltopology and can define their won virtual network topologyOVXwill assist its users to accomplish the mapping of virtualnetwork topology to physical network topology

The conventional offline benchmark planning is nolonger suitable to satisfy 5G 120583Osrsquo efficient utilization ofresources and demands of sharing and the full utilization ofcomputing resources to respond to the complicated situationsof different demands to determine the direction of the trafficflow Machine learning is on the list of the most potentialcandidates [36] It provides a much stronger and complicateddecision-making ability Decision tree is one of the mostpopularmachine learning algorithms used all alongDecisiontree on the other hand is highly efficient supervised learningmodel that is ideal for the prediction of classification andregression data type such as ID3 and C45 [37 38] C45is the refined version of feature selection in ID3 algorithmthat can be used to process continual class label and theproblem of missing data in training data It can also correct

the problem of overfitting of the decision tree by means ofpruning As for the application of route choosing and trafficflow identification in regard ofmobile edge network [39] usesthe characters of SDN and combines them with decision treealgorithm to categorize traffic flow allowing the controller tochoose an access point from the local network that is suitableto be connected to a mobile device according to the levelof congestion of the backhaul route Reference[40] proposesa model called AMPS that combines SDN and machinelearningThe model is capable of categorizing the differencesof each traffic flow by means of learning packet identificationand then chooses the best route of transmission for the trafficflow to achieve the automation of the optimal selection ofrouting path

3 5G 120583O Design Pattern andDTBFR Mechanism

The interaction between SDN and NFV enables the 5Gnetwork to build a system infrastructure in an abstractmanner and further increase the flexibility of the networkallows the vertical system to be divided into multiple slicedconstructor blocks and builds a network that is connectableprogrammed and virtual The operator can advance to usenetwork slicing technology to virtualize networks and asthe aforesaid paragraphs flexibly divide a physical networkinto multiple independent and segregated 120583O networksaccording to different usage scenarios As for the designof the infrastructure this paper uses SDN and NFV tech-nologies as the base and combines network slicing andtunneling technologies to build a network infrastructure for120583Os The infrastructure allows users of different 120583Os tobe concatenated via tunneling technology and then realizesthe rapid connection of network to effectively enhance theinterconnectivity of networks

The proposed 120583O design pattern is shown in Figure 3Network slicing technology realizes the logical partition of120583Osrsquo networks via OpenVirtex The connection between the


core network and the 120583O is carried out by the tunnel built bySDNborder gateway (BG)The Internet cloud data center andthe micro data center are constructed by NFV infrastructuredefined by ETSI to save the cost of equipment investmentThemission of SDN controller is to use OpenFlow to constructa passageway between BG at the edge of the core networkand BG at the edge of the 120583Orsquos network Once the SDNcontroller begins executing the matching and concatenationbetween network passageways the 120583Orsquos network can build aconnection via the tunnel and the main core network The120583Ocan continue finishing the construction of virtual networkvia OpenVirtexrsquos virtual slicing technology allowing users ofthe 120583O to gain access to the data on the nearby micro datacenter The users can also via tunnels connect to the clouddata center on the Internet to access the network serviceof particular application The proposed architecture patterncombines network slicing and tunneling to realize the com-munication model for 120583Os and further integrates bandwidthcontrolling technology to be applied to the communicationbandwidth application of the 120583Orsquos networkThis will increasethe utilization ratio of network resources and the efficiency oftraffic flow and result in a better quality of experience (QoE)for the network users

In response to the demand of 120583Orsquos network resourcedistribution that allows users to gain access of nearby networkresources the paper proposes a DTBFR mechanism for 120583Oand uses decision tree theory to serve as the reference in thedetermining the SDN traffic flows direction Decision treealgorithms have been used for solving predictive analyticsproblems in the past few years Decision tree is a supervisedmachine learning model with simple process intuition andhigh execution efficiency Decision tree algorithm is mainlyused to conduct systematic result and integration of datasetand to find the special class and label relation during thedecision-making process Compared with other ML modelsexecution speed is a major advantage To quantify uncertaininformation and dataset the paper uses entropy as a methodto gauge uncertainty and level of chaos in (1) Entropy is ameasure of the impurity in a collection of training examplesThe entropy increases with the increase in uncertainty orrandomness and decreases with a decrease in uncertaintyor randomness The production of information comes withuncertainty and entropy can gauge it according to its proba-bility of occurrence [41] With higher probability the chaos ismore likely to occur and the uncertainty is low while on thecontrary the uncertainty is high

119864119899119905119903119900119901119910 (119904) = 119888sum119894=1

119875119894 log2 119875119894 (1)

Decision trees algorithm uses feature selection to guidethe decision of the most useful attributes Different featureselection criteria result in different types of decision trees ID3uses entropy and information gain to construct a decisiontree Information gain is used to decide which feature tosplit on at each step in building the tree Information gainmeasures the expected reduction in entropy by partitioningthe examples according to an attribute Calculations of ID3information gain are shown in (2) The information gains

criteria to measure the strength of the association betweenan attribute and class Symbol S is the target class and SymbolA represents the class label C45 is an extension of ID3 whichis a similar tree generation algorithm C45 uses gain ratioby splitting the training sets based on its test attributes Thegain ratio takes number and size of branches into accountwhen choosing an attribute It corrects the information gainby taking the intrinsic information of a split into accountThe split information can be defined as shown in (3) Thegain ration can be defined as shown in (4) The flowchart ofdecision tree construction is shown in Figure 4

119866119886119894119899 (119878119860) = 119864119899119905119903119900119901119910 (119878)minus Vsum119895=1

1003816100381610038161003816100381611987811989510038161003816100381610038161003816|119878| 119864119899119905119903119900119901119910 (119878119895) (2)

119878119901119897119894119905119868119899119891119900 (119878 119860) = minus 119888sum119894=1

10038161003816100381610038161199041198941003816100381610038161003816|119878| log2 (10038161003816100381610038161198781198941003816100381610038161003816|119878| ) (3)

119866119886119894119899119877119886119905119894119900 (119878 119860) = 119866119886119894119899 (119878 119860)119878119901119897119894119905119890119868119899119891119900 (119878 119860) (4)

The dataset used by the decision tree consists ofboth class labels and decision result Training data col-lects user application traffic primarily through SDN con-troller The dataset includes userrsquos network area serverrsquosCPUmemory and traffic oriented result The data stor-age format is represented as [user area s1 cpu s2 cpus3 cpu s1 mem s2 mem s3 mem result] For example thedataset [2 490 356 60 290 130 363 2] corresponds tothe class label [120583O s0 cpu s0 memory s1 cpu s1 mems2 cpu s2 mem result] Training data is a certain percentageof an overall dataset along with a testing set The more com-plete training data makes classification easier The duplicatevalues and divorced values must be removed first from thedataset Duplicate values refer the same values for each classlabel in the dataset Divorced values are calculated from theaverage value of the class label When the value of the classlabel is greater than the average of plusmn5 the dataset will beremoved from dataset

After completing training on data collection we canstart to establish the node of the decision tree Calculatethe Gain(S A) SplitInfo(S A) and Gain ratio(S A) fromthe class labels of the dataset Pick the best Gain ratio (SA) from the list and save it to BestFeature as the root ofthe decision tree The remaining unselected class labels areadded to Unselected labels The execution of the recursionbegins when the decision tree node is established Pruningis a technique in machine learning that reduces the size ofdecision trees by pruning the tree based on the statisticalconfidence estimates When the decision tree is created loadthe training data and start traverse tree node from root toleaf We adopted the pessimistic pruning strategy proposedused in C45 to avoid the need of pruning set and continuitycorrection to error rate at each node

In this paper the calculation of continuous class labelsand discrete class labels is different Discrete data is the typeof data that has clear spaces between values while continuous


Table 1 Example of continuous class labels

Number 1 2 3 4 5 6 7 8 9 10 11 12 13 14CPU 854 903 784 805 805 752 660 903 752 805 758 903 752 903S No No Yes No Yes No Yes No Yes Yes Yes Yes Yes Yes

Start Training data

Calculate of Gain ratio and Split info

from the class labels of the dataset

Create atree node byBestFeature

Calculate the Gain ratio and Split info of

Unselected_labels

Create a internal node tree byBestFeature

If all the attributesalready calculated

maximum Gain ratio

Decision Tree createdTraining data

Traverse the treenode fromroot to leaf

Calculate error_subtree var(subtree)

error_leaf

NO

YES

Pruning

YES

If all tree nodesalready traversed

Pruned Tree created

End

NO

YES

NO

If error_subtree-var(subtree) gt

error_leaf

Figure 4 The decision tree construction flowchart

data is data that falls on a continuous sequence Table 1 showsthe example of continuous class label The second columnshows the serversrsquo CPU usage in 120583O The third column Srepresents the output of the decision In this example we startwith number 66 to calculate the gain ratioThe dataset greaterthan 660 and the dataset less than or equal to 660 use (1) tocompute the entropy value as 0 and 096 respectively Thenthe results are calculated by (2) and (3) Dividing theGain (S660) by the (3) yields the gain ratio value of 0129 Table 2shows the calculation results and the class label CPU 660gets the maximum gain ratio value in this example Basedon the above calculation the best value is chosen in eachround to serve as the node of the tree until the decision treeis established

Based on the results of Table 2 the decision tree isconstructed as shown in Figure 5 In order to correct theoverlearning problem caused by the impure information inthe decision tree we adopted the pessimistic error pruningstrategy to recursively estimate the error rate of the samplenodes covered by each internal node Pessimistic error prun-ing is based on error estimates derived from the training data

If the internal nodes error rate is lower than the estimatedvalue it will be replaced by the largest number of leaf nodesin the subtree

Assume that an internal node of the tree covers N sampleswith E errors then the error rate of the leaf nodes is givenby (5) This penalty factor is 05 Pessimistic error pruningadds a constant to the training error of a subtree by assumingthat each leaf automatically classifies a certain fraction of aninstance incorrectlyThis fraction is taken to be 12 divided bythe total number of instances covered by the leaf and is calleda continuity correction in statistics

119864 (119897119890119886119891 119890119903119903119900119903 119903119886119905119890) = (119864 + 05)119873 (5)

119864 (119904119906119887119905119903119890119890 119890119903119903119900119903 119903119886119905119890) = (119864 + 05 lowast 119897119890119886119891)119873 (6)

After pruning the internal node becomes a leaf node andthe false positive number also need to add a penalty factorThe error rate of subtree is given by (6) Equations (7) and (8)show the false positive number and the standard deviation of


Table 2 Calculation result of the continuous class labels

Value 660 752 758 784 805 854 903interval ≦ gt ≦ gt ≦ gt ≦ gt ≦ gt ≦ gt ≦ gtYes 1 8 3 6 4 5 5 4 7 2 7 2 9 0No 0 5 1 4 1 4 1 4 2 3 3 2 5 0Entropy 0 096 081 097 072 099 065 1 076 097 088 1 065 0Info(SA) 0891 092 089 085 0838 091 0651Gain 0048 0015 0045 009 0102 002 0Split Info 0371 0863 094 098 094 0863 0651Gain ratio 0129 0017 0047 0091 0108 0023 0

s2_mem

s2_cpu s2_cpu

s1_mem

S1_cpu

s0_cpu

s0_mem s0_mem

1 0

0

1 0 2

S1_mem

0

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 2 0 0 2

0

1 0 1 0

1

11 0

746 gt 746 lt=

500 lt=

580 lt=

1 0 1 0 1 0

0 2 0 21 0 1 0 22

1 0 2

1 0 580 gt

1 0 2580 gt

620 gt 620 lt=

602 lt=

580 lt=758 gt 758 lt=

742 gt 742 lt=

630 lt= 630 gt

668 gt668 lt= 668 gt668 lt=

680 gt680 lt=

728 lt=728 gt

660 lt= 660 gt 500 gt

602 gt

O

O

OOO

O O O O

O

Figure 5 An example of established decision tree

the subtree The false positive number of leaf node is shownin (9)

119864 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (7)

V119886119903 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905) = radic119873 lowast 119890 lowast (1 minus 119890) (8)

119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (9)

119864 (119904119906119887 119890119903119903119900119903 119888119900119906119899119905) minus V119886119903 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905)gt 119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) (10)

As shown in (10) if the number of misclassified subtreesminus the standard error of the error rate is still greaterthan the number of false positives corresponding to the leafnode then the pruning will occur and the subtree will beremoved from the tree Until all nodes in all decision trees

have been visited the pruning process is completed Thepruning process is finished when all the nodes in all thedecision trees are visited Figure 6 shows the pruned decisiontree

The proposed DTBFR mechanism utilizes the SDN tech-nology to collect the network traffic information The calcu-lation results of decision tree algorithm can provide instruc-tions for SDN controller to guide user traffic flows to theservices in the nearby 5G 120583O micro data center The DTBFRmechanism combined with 5G SDNNFV programmablenetwork architecture can improve service efficiency and saveavailable bandwidth resources

4 Experiment Results

The purpose of this experiment is to verify 120583Orsquos communica-tion infrastructure pattern during operationThe experiment


s2_mem

s2_cpu

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 0 2 0 2

1

1

746 gt746 lt=

500 lt=

580 lt=

0 2 1 01 0 2 0 22

580 gt

1 0 2680 gt

620 gt 620 lt=

602 lt=

680 lt=758 gt 758 lt=

500 gt

602 gt

0

O O O O

O

Figure 6 The pruned decision tree

condition consists of multiple local area networks controlledby the 120583O and central tunnel controllers In our evaluationwe simulated a network with tree-like topology in mininetand measured the overhead caused by network view division[42] Mininet is used to simulate 120583Orsquos network and tunnelscan be built among different 120583O areas for quick exchangeof information The experiment comprises two hypervisorsone OpenVirtex and multiple SDN controllers shown inFigure 7 The topology is constructed by the mininet inVM in the hypervisor and the 120583O areas are concatenatedby tunneling technology The OpenVirtex functions as thesystem management system The IT staff can constructnetwork slicing for the system via network interface and theback-end server of the network interface will analyse theoperating information before interpreting it into OpenVirtexto produce commands for network slicingThe webpage usessocket and slice manager to transmit information After theslice manager receives the information it will continue touse OpenVirtexrsquos API to build network slicing for the mana-ger

Network slicing has a wide range of applications accord-ing to the cohorts of users the types of application servicesor different demands for resources The paper uses region-alization service as the benchmark for network slicing andconnects homogeneous regionalization service servers withtunneling technology to form a complete network slicing ofregionalization service A C45 decision tree model is builtbased on the mechanism proposed by Ross Quinlan to serveas the reference for determination of 5G 120583Orsquos traffic flowdirection and allows the usersrsquo traffic flow to be directed toa closer 120583O access service to reduce the network latency inpacket transmission

In order to verify 120583O 0 and 120583O 1rsquos virtual communica-tion service infrastructures the experiment uses hypervisorto stimulate the scenario of adjacent 120583O communicationThe infrastructure of the experiment consists of one Xenhypervisor and two Ubuntu 1604 virtual machines Mininetis used to virtualize S1 for VM1 add GRE interface and setthe remote IP as 101010126 which is the IP of VM2 VM2has the same setting with the only difference that its remoteIP is set as 101010116 which is the IP of VM1 Both VM1and VM2rsquos control plane are set as OpenVirtex and OVXwill simultaneously finish logical network slicing when theslicing is built As shown in Figure 8 the TCP bandwidthtest is conducted by Iperf in mono-hypervisor Iperf is anetwork benchmarking tool which is used to measure thethroughput of the network carrying UDP and TCP data [43]The average bandwidth of GRE tunnels is 165 Gbps andthe average bandwidth of VxLAN tunnels is 163 Gbps Theresults of the experiment are shown in Figure 9 As shownin Figure 10 logical network slicing infrastructure is builtby the two hypervisors Hypervisor 1 and Hypervisor 2 areseparated physicallyThe results of the experiment are shownin Figure 11 In this case the average bandwidth of GREtunnels is 915 Mbps and the average bandwidth of VxLANtunnels is 908 Mbps

Judging from the results we can see that even thoughtunnels between the two VMs in mono-hypervisor aresuccessfully built the packets are unable to actually leavehypervisorrsquos network card but exchange memory insteadThis is because both VMs are located in the same hypervisorIn the two hypervisors scenario tunnels are successfully builtas well and the packets actually reach each otherrsquos hypervisornetwork cards to carry out the exchange This is attributed


Xen Server2X

eth0

SW8

SW2SW7GRE1

uo0 network

100080

101010130

eth0

SW3

SW9

GRE

uo1 network 1000160

101010116

eth0

SW4

SW5

SW1GRE

core network

10004

10101012610007GRE2

SW6

10003

1000510001

streaming server

streaming server

broadcaster

watcher1000102

1000253streaming server

Xen Server1

Xen Server1

OpenVirtex

Slice1 Controller Slice2 Controller Slice3 Controller

Physical LinkControl Link

IT operator

Web GUI

Slice manager

Tunnel

Tunnel

Figure 7 The experiment environment

Hypervisor

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0

101010116 101010126

Tunnel

s2

gre

s1

gre

Figure 8 Average bandwidth test in a signal hypervisor scenario

15155

16165

17175

0 5 10 15 20 25 30 35 40 45 50 55 60

band

wid

th (G

bps)

time (sec)

GREVxLAN

Figure 9 The experiment result in a signal hypervisor


Hypervisor 1

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0101010116 101010126

Tunnel

s2gre

s1

gre

Hypervisor 2

Figure 10 Average bandwidth test in two hypervisors scenario

860880900920940960

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60band

wid

th (M

bps)

time (sec)

VxLAN

Figure 11 The experiment result in cross-hypervisors

to that fact that a bridge of communication has been builtby the tunnels and the packets are repackaged It is thereforedetermined that the two hypervisors scenario requires lessbandwidth than mono-hypervisor

As for the scenario of nonadjacent 120583O communica-tion we replace the experiment infrastructure with two-tiermininet to simulate the whole environment and constructa virtual experiment infrastructure in different domains Asshown in Figure 12 in the first tier we use the mininetto construct topology in different domains and connectdomain 1 and domain 2 via the OpenVSwitch (OVS) of themininet Domain 1 and Domain 2 can also construct theirown topology via the mininet The topology of the mininetin the first tier uses OpenVirtex to conduct the networkslicing for tenants The slices are composed of tunnel andnontunnel and the slices are subject to the management ofdifferent controllers For example the tunnel slice is subjectto tunnel controller and so forth The tunnel network slicinguses the exclusive network slices built by the tunnelingtechnology The experiment uses two mainstream tunnelingtechnologies GRE and VxLAN to conduct the comparisonThe experiment result is shown in Figure 13 The experimentsets the bandwidth of the link in the topology at 100Mbytesand the major purpose of the experiment is to test thetransmission efficiency of the two tunneling technologiesGRE tunnel package belongs to the third tier routing VxLANuses UDP to pack the packet The average bandwidth ofGRE is around 737 Mbytes and the average bandwidth of

VxLAN is around 348MbytesThe results clearly suggest thatVxLANrsquos packet transmission efficiency is not even half ofthat of GRErsquosThe cause behind this is because of the differentways of packet transmission UDP is by itself an unreliableway of transmission and the packet is even more likely to belost with theOpenVirtex serving as the intermediary tierThismight explain the poor performance in bandwidth

To evaluate the efficiency and the performance theDTBFRmechanism proposed in this paper is compared withthe Minimize Loading First (MLF) mechanism The MLF isa greedy algorithm that can be used to find the server withthe lowest load and direct the traffic to the server via SDNcontroller

In order to make users of 120583O gain access to the nearbymicro data center or link to the Internet data center for accessto particular application of network services via tunnelsthree indicators are available to determine the quality of thedecision-making of traffic flow redirection 120583O num repre-sents the number of accessed services in 120583O Near 120583O numrepresents the number of times when the traffic flow hasto access services in nearby 120583O because the servers in 120583Oare all busy Datacenter num represents the number of timesrecorded when the traffic flow has to access remote datacenter because nearby 120583Os are all busy The results of theexperiment are shown in Figure 14 The value of 120583O num ofthe proposed DTBFR mechanism is apparently higher thanthat ofMLFmechanismThismeans that DTBFRmechanismis able to prioritize the userrsquos traffic flow according to theservers in 120583Orsquos area If the server is busy DTBFR mechanismwill consider accessing services in nearby 120583O and becausethe number of HTTP requests varies (11) is used to calculatethe redirection ratio

Ratio = (120583O numnear 120583O numDatacenter num)Total number of HTTP requests

(11)

Under the same experiment conditions the decision-making of traffic flow direction of DTBFR mechanism is asfollows 681 of the traffic flow will be directed to the 120583O inthe current area to access services 287 of the traffic flowwill be directed to the nearby 120583O to access services only32 of the traffic will go to the data center to access services


eth1eth0

SW1

SW4

SW2

SW3

BGGRE

eth1eth0

SW1

SW4

SW2

SW3

BGGRE

Domain2 NetworkDomain1 Network

1000100 101010170 1000200 10101046

Tunnel Slice

Non-Tunnel Slice

192003192001

101010174

SDN Controller

Openvirtex

Xenserver

TunnelController

Non-TunnelController

10001

10003

Figure 12 Nonadjacent 120583O communication scenario

0102030405060708090

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60

Band

wid

th (M

byte

s)

time (sec)

VxLAN

Figure 13 The comparison of GRE and VxLAN experiment resultin nonadjacent 120583O communication scenario

MLF algorithm is a directing method based on server loadwhich directs 298 of the traffic flow to the 120583O in the currentarea to access services 327 of the traffic flow to the nearby120583O to access services and 375 of the traffic flow to theInternet data center to access services The 120583Orsquos accessedservices indicator in the area shows that DTBFR mechanism

takes up 681whileMLF algorithm takes up only 298Theexperiment verifies that DTBFR mechanismrsquos performanceon traffic redirection is relatively good

In summary the design pattern of 5G micro operatorcombines network slicing and tunneling technologies torealize the communication model for 120583O although networkslicing holds much promise for 5G but not without its shareof hurdles The traditional network architecture will needto be redesigned to enable network slicing Interoperabilityshould be tested in order to ensure network slicing worksas expected in the 5G network In terms of tunnelingthe inefficient packet overhead may have a negative effecton the performance of the network For the purpose ofmicro operator to providing regional services the DTBFRmechanism allows different 120583Os to share network resourcesand improves data processing by directing the usersrsquo trafficflow to a closer 120583O for reducing the network transmissionlatency building the foundations of ultrareliable and low-latency communications (URLLC) in 5G Conclusively the120583O design pattern provided in this paper helps to reduce thecomputing burden of the traditional cloud network architec-ture improves the load capacity of the cloud network and theservers and is also able to be used as development basis for


near_O_num O_num Datacenter_numDTBFR 13346 31701 1441MLF 28185 25707 32109DTBFR ratio 0287 0681 003MLF ratio 0327 029 037

0

5000

10000

15000

20000

25000

30000

35000

00102030405060708

num

ber o

f req

uest

s

Ratio

of r

edir

ectio

n

DTBFR

MLF

DTBFR ratio

MLF ratio

Figure 14 Comparison with experiment results of DTBFR and MLF

integration and innovation application of the Cloudy-EdgeComputing achieved by 5G

5 Conclusions

This paper uses SDN and NFV technologies as the baseand combines network slicing and tunneling technologies tocome up with a network infrastructure pattern for 120583Os Thepattern allows users of different 120583Os to be concatenated viatunneling technology and then realizes the rapid connectionof network to effectively enhance the interconnectivity ofnetworks In order to meet the demand of regional serviceof micro operator this paper proposes DTBFR mechanismwhich uses decision tree theory as the basis of SDN-basedtraffic decision-making Under the same experiment condi-tions the decision-making of traffic flow direction of DTBFRmechanism is as follows 681 of the traffic flow will bedirected to the120583O in the current area to access services 287of the traffic flow will be directed to the nearby 120583O to accessservices only 32 of the traffic will go to the Internet clouddata center to access services The features deployed by theregional service of the micro operator can effectively reducethe burden of the data center on the Internet and acceleratethe development of the edge computing service in the future5G network

Future work is needed to create an edge computingservice model for the wide-spread adoption of the 120583O designpattern especially the integration of the edge computingtechnologies into the 5G 120583O deployment path by improvingmanagement efficiency and provides service on-demandapplication for micro operator customers The resource esti-mation tasking scheduling and lightweight virtual networkfunction configuration techniques are the primary researchdirections

Data Availability

The data used to support the findings of this study areincluded within the article

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Acknowledgments

Thework described in this paper was supported in part by theMinistry of Science and Technology of the Republic of China(Project nos 104-2221-E-008 -039 105-2221-E-008-071 107-2623-E-008-002 and 107-2636-E-003-001)

References

[1] X Huang Y Li S Tang and Q Chen ldquoCoexistence ofcognitive small cell and WiFi system a traffic balancing dual-access resource allocation schemerdquo Wireless Communicationsand Mobile Computing vol 2018 Article ID 4092681 17 pages2018

[2] A Raschella F Bouhafs G C Deepak and M Mackay ldquoQoSaware radio access technology selection framework in hetero-geneous networks using SDNrdquo Journal of Communications andNetworks vol 19 no 6 pp 577ndash586 2017

[3] H Beyranvand M Levesque M Maier J A Salehi C Verik-oukis and D Tipper ldquoToward 5G FiWi enhanced LTE-Ahetnets with reliable low-latency fiber backhaul sharing andWiFi offloadingrdquo IEEEACM Transactions on Networking vol25 no 2 pp 690ndash707 2017

[4] F Z Yousaf M Bredel S Schaller and F Schneider ldquoNFV andSDN-Key technology enablers for 5G networksrdquo IEEE Journalon Selected Areas in Communications vol 35 no 11 pp 2468ndash2478 2017

[5] C W Tseng P H Lai B S Huang L D Chou and M CWu ldquoNFV deployment strategies in SDN networkrdquo Interna-tional Journal of High Performance Computing and Networking(IJHPCN) 2018 Available httpwwwindersciencecominfoingeneralforthcomingphpjcode=ijhpcn

[6] E J Kitindi S Fu Y Jia A Kabir and Y Wang ldquoWirelessnetwork virtualization with SDN and C-RAN for 5G networksrequirements opportunities and challengesrdquo IEEE Access vol5 pp 19099ndash19115 2017


[7] uO5G ldquoMicro operator concept for boosting local servicedelivery in 5Grdquo Available httpwwwoulufiuo5g

[8] M Matinmikko M Latva-aho P Ahokangas S Yrjola and TKoivumaki ldquoMicro operators to boost local service delivery in5GrdquoWireless Personal Communications vol 95 no 1 pp 69ndash822017

[9] A Prasad Z Li S Holtmanns and M A Uusitalo ldquo5G micro-operator networks mdash A key enabler for new verticals andmarketsrdquo in Proceedings of the 2017 25th TelecommunicationForum (TELFOR) pp 1ndash4 Belgrade Serbia November 2017

[10] J S Walia H Hammainen and M Matinmikko ldquo5G Micro-operators for the future campus A techno-economic studyrdquoin Proceedings of the 2017 Internet of Things - Business ModelsUsers and Networks pp 1ndash8 Copenhagen Denmark Novem-ber 2017

[11] M Matinmikko-Blue and M Latva-aho ldquoMicro operatorsaccelerating 5G deploymentrdquo in Proceedings of the 2017 IEEEInternational Conference on Industrial and Information Systems(ICIIS) pp 1ndash5 Peradeniya Sri Lanka December 2017

[12] K B Manosha M Matinmikko-Blue and M Latva-aholdquoFramework for spectrum authorization elements and its appli-cation to 5Gmicro-operatorsrdquo inProceedings of the 2017 Internetof Things - Business Models Users and Networks pp 1ndash8Copenhagen Denmark November 2017

[13] T Sanguanpuak S Guruacharya E Hossain N Rajatheva andM Latva-aho ldquoOn spectrum sharing amongmicro-operators in5Grdquo in Proceedings of the 2017 European Conference onNetworksand Communications (EuCNC) pp 1ndash6 Oulu Finland June2017

[14] P Ahokangas S Moqaddamerad and M Matinmikko ldquoFuturemicro operators business models in 5Grdquo The Business andManagement Review vol 7 no 5 pp 143ndash149 2016

[15] P Mach and Z Becvar ldquoMobile Edge Computing A Survey onArchitecture and Computation Offloadingrdquo IEEE Communica-tions Surveys amp Tutorials vol 19 no 3 pp 1628ndash1656 2017

[16] G Li J Song J Wu and J Wang ldquoMethod of resourceestimation based onQoS in edge computingrdquoWireless Commu-nications and Mobile Computing vol 2018 9 pages 2018

[17] M A Lema A Laya T Mahmoodi et al ldquoBusiness caseand technology analysis for 5G low latency applicationsrdquo IEEEAccess pp 5917ndash5935 2017

[18] P Kiss A Reale C J Ferrari and Z Istenes ldquoDeployment ofIoT applications on 5G edgerdquo in Proceedings of the 2018 IEEEInternational Conference on Future IoT Technologies (FutureIoT) pp 1ndash9 Eger Hungary January 2018

[19] M Matinmikko A Roivainen M Latva-aho and K HiltunenldquoInterference study of micro licensing for 5g micro operatorsmall cell deploymentsrdquo in Cognitive Radio Oriented Wire-less Networks vol 228 of Lecture Notes of the Institute forComputer Sciences Social Informatics and TelecommunicationsEngineering pp 264ndash275 Springer International PublishingSwitzerland 2018

[20] L Falconetti R Karaki and S Corroy ldquoPractical energy-aware cell association for small cell deploymentrdquo WirelessCommunications and Mobile Computing vol 16 no 16 pp2436ndash2448 2016

[21] T Zahid X Hei We Cheng A Ahmad and P Maruf ldquoOnthe tradeoff between performance and programmability forsoftware definedWiFi networksrdquoWireless Communications andMobile Computing vol 2018 Article ID 1083575 pp 1ndash12 2018

[22] P Ahokangas M Matinmikko S Yrjola and I Atkova ldquoDis-ruptive revenuemodels for futuremicro operator drivenmobile

business ecosystemrdquo in Proceedings ofThe 24th Nordic Academyof Management Conference (NFF) pp 23ndash25 Bodo Norway2017

[23] MG Kibria G P Villardi KNguyenW-S Liao K Ishizu andF Kojima ldquoShared spectrum access communications a neutralhost micro operator approachrdquo IEEE Journal on Selected Areasin Communications vol 35 no 8 pp 1741ndash1753 2017

[24] ldquoMicro operators for vertical specific service delivery in 5GrdquoAvailable httpwww5gsummitorgberlindocsslidesMatti-Latva-Ahopdf

[25] A Basta A Blenk K Hoffmann H J Morper M HoffmannandW Kellerer ldquoTowards a cost optimal design for a 5Gmobilecore network based on SDN and NFVrdquo IEEE Transactions onNetwork and Service Management vol 14 no 4 pp 1061ndash10752017

[26] C Bouras A Kollia and A Papazois ldquoSDN and NFV in5G Advancements and challengesrdquo in Proceedings of the 20thConference on Innovations in Clouds Internet and NetworksICIN 2017 pp 107ndash111 France March 2017

[27] ONF ldquoSDNarchitecturerdquoAvailable httpswwwopennetwork-ingorgimagesstoriesdownloadssdnresourcestechnical-re-portsTRSDNARCH1006062014pdf

[28] ETSI ldquoNetwork Functions Virtualization (NFV) architecturalframeworkrdquo ETSI Ind Specification Group Sophia AntipolisCedex France 2014

[29] Q Jia R Xie T Huang J Liu and Y Liu ldquoEfficient cachingresource allocation for network slicing in 5G core networkrdquo IETCommunications vol 11 no 18 pp 2792ndash2799 2017

[30] X Li M Samaka H A Chan et al ldquoNetwork slicing for 5Gchallenges and opportunitiesrdquo IEEE Internet Computing vol 21no 5 pp 20ndash27 2017

[31] K Samdanis X Costa-Perez and V Sciancalepore ldquoFromnetwork sharing tomulti-tenancyThe5Gnetwork slice brokerrdquoIEEE Communications Magazine vol 54 no 7 pp 32ndash39 2016

[32] NGMN Alliance ldquo5G White Paperrdquo Available httpswwwngmnorg5g-white-paperhtml

[33] NOKIA ldquoDynamic end-to-end network slicing for 5G WhitePaperrdquo Available httpwwwhitbmehusimjakabedulitr5GNOKIA dynamic network slicing WPpdf

[34] S Min S Kim J Lee B Kim W Hong and J Kong ldquoImple-mentation of an OpenFlow network virtualization for multi-controller environmentrdquo in Proceedings of the 14th InternationalConference on Advanced Communication Technology ICACT2012 pp 589ndash592 Republic of Korea February 2012

[35] A Al-Shabibi M De Leenheer M Gerola et al ldquoOpenVirteXMake your virtual SDNs programmablerdquo in Proceedings of the3rd ACM SIGCOMM 2014 Workshop on Hot Topics in SoftwareDefinedNetworkingHotSDN2014 pp 25ndash30 Chicago ILUSAAugust 2014

[36] E Alpaydin Introduction to Machine Learning The MIT PressCambridge Mass London England 2nd edition 2009

[37] B Hssina A Merbouha H Ezzikouri and M Erritali ldquoAcomparative study of decision tree ID3 and C45rdquo InternationalJournal of Advanced Computer Science and Applications vol 4no 2 2014 Special Issue on Advances in Vehicular Ad HocNetworking and Applications 2014

[38] R C Barros M P Basgalupp A C P L F De Carvalhoand A A Freitas ldquoA survey of evolutionary algorithms fordecision-tree inductionrdquo IEEE Transactions on Systems Manand Cybernetics Part C Applications and Reviews vol 42 no 3pp 291ndash312 2012


[39] D Lee and C S Hong ldquoAccess point selection algorithmfor providing optimal AP in SDN-based wireless networkrdquo inProceedings of the 2017 19th Asia-PacificNetworkOperations andManagement Symposium (APNOMS) pp 362ndash365 Seoul SouthKorea September 2017

[40] S T V Pasca S S P Kodali and K Kataoka ldquoAMPS Appli-cation aware multipath flow routing using machine learning inSDNrdquo in Proceedings of the Twenty-third National Conference onCommunications (NCC) pp 1ndash6 Chennai India 2017

[41] ldquoEntropyrdquoAvailable httpsenwikipediaorgwikiEntropy (in-formation theory)

[42] ldquoMininetrdquo Available httpmininetorg[43] ldquoiPerfrdquo Available httpsiperffr

International Journal of

AerospaceEngineeringHindawiwwwhindawicom Volume 2018

RoboticsJournal of

Hindawiwwwhindawicom Volume 2018


Active and Passive Electronic Components

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawiwwwhindawicom

Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Control Scienceand Engineering

Journal of



Journal ofEngineeringVolume 2018

SensorsJournal of



RotatingMachinery


Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation


Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia

Submit your manuscripts atwwwhindawicom


Micro OperatorInternet

3G

LTEMacrocell

Micro DataCenter

Firewall

Firewall

Data Center

VoIPServer

CloudVirtualizedServers

HTTPServer

StreamingServer Application

Server

DSNAAAServer

SDNOpenFlowController

WiFi IEEE80211

abgnp

IP Tunnel

Figure 1 Micro operator network architecture

users The 120583O is originated from Oulu University in Finlandthat consists of three features [14] (1) Mobile devices accesslocal network specific service (2) spatially confined areawith specific service (3) dependent on appropriate availablenetwork resources Facilities like hospitals schools largeconference sports venues shopping malls hypermarketsand factories can use local area network service to satisfyusersrsquo needs for all sorts of application services

Restricted by resources and space 120583O has the characterof regionalization service that allows it to provide undercircumstances where hardware infrastructures are scarceand the resources are limited different regional servicesaccording to different domains making it possible for usersof mobile devices to gain access to services in the nearbyregions Apart from reducing the consumption of bandwidthresources by offering nearby network services this kind oflocalization service can also transfer applications data andthe computing of services from nodes in the data centeron cloud to edge nodes in logical LAN to be processedand realize the developing environment for edge computing[15 16] reduce network latency and satisfy 5G demands foramelioration of latency

The purpose of this paper is to discuss the 120583O designpattern in 5G SDNNFV network and establishes an experi-mental environment to evaluate the decision tree based trafficflow redirectionmechanismTo realize the customized exper-iment environment the network functions virtualization istaken into consideration in our research Our contributionsare listed as follows

(1) A120583Odesign pattern is constructed based on SDNandNFV that combines network slicing and tunnelingtechnologies to build a network infrastructure for120583Os

(2) A DTBFR mechanism for 120583O is proposed and thedecision tree theory is utilized to serve as the referencein the determining the SDN traffic flows direction

(3) We established a 120583O environment for validation ofthe DTBFR mechanism and 120583Os communicationexperiments

The rest of the paper is organized as follows in Section 2the background and related works are addressed Section 3describes the 120583O design pattern and DTBFR mechanismSection 4 presents the experiment results The last sectionconcludes this paper

2 Background and Related Works

The development of 5G application services will be largelyon the Internet of Things (IoT) and encourages the newcommunication market to move toward a more vertical-subdivision one [17 18] The result will be the formation ofdifferent emerging application service scenarios and morediverse demands for networks However as far as telecom-munication is concerned despite the fact that there is nowa globally agreed requirement on IoT constructed by 5G oncharacters such as transmitting speed capacity coverage andsecurity there is still room for the business model of 5Gto improve itself 5G mobile broadband network puts theaccent on small cellsbase stations how to amplify indoorcoverage providing faster services for users and reducingthe latency in network transmission all of which pose agreat challenge for telecommunication operators [19ndash21]Furthermore the appearance of MVNOs has brought newopportunities for development for MNOs who have not yetgained licenses to operate mobile communication business




Micro Operator








(O)












Core network

BG

BGOVS

OVS

OVSOVS

BG

Micro Operator

Datacenter

BGBG

3G

4GLTE

OpenVirtex

tunnel

tunnel tunnel

tunnel

tunnel

NFVI

NFVI


Slice 3

Slice 2

Slice 1



tunnel











119864119899119905119903119900119901119910 (119904) = 119888sum119894=1

119875119894 log2 119875119894 (1)



119866119886119894119899 (119878119860) = 119864119899119905119903119900119901119910 (119878)minus Vsum119895=1

1003816100381610038161003816100381611987811989510038161003816100381610038161003816|119878| 119864119899119905119903119900119901119910 (119878119895) (2)

119878119901119897119894119905119868119899119891119900 (119878 119860) = minus 119888sum119894=1

10038161003816100381610038161199041198941003816100381610038161003816|119878| log2 (10038161003816100381610038161198781198941003816100381610038161003816|119878| ) (3)

119866119886119894119899119877119886119905119894119900 (119878 119860) = 119866119886119894119899 (119878 119860)119878119901119897119894119905119890119868119899119891119900 (119878 119860) (4)







Start Training data





Unselected_labels



maximum Gain ratio




error_leaf

NO

YES

Pruning

YES


Pruned Tree created

End

NO

YES

NO


error_leaf






119864 (119897119890119886119891 119890119903119903119900119903 119903119886119905119890) = (119864 + 05)119873 (5)

119864 (119904119906119887119905119903119890119890 119890119903119903119900119903 119903119886119905119890) = (119864 + 05 lowast 119897119890119886119891)119873 (6)





s2_mem

s2_cpu s2_cpu

s1_mem

S1_cpu

s0_cpu

s0_mem s0_mem

1 0

0

1 0 2

S1_mem

0

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 2 0 0 2

0

1 0 1 0

1

11 0

746 gt 746 lt=

500 lt=

580 lt=

1 0 1 0 1 0

0 2 0 21 0 1 0 22

1 0 2

1 0 580 gt

1 0 2580 gt

620 gt 620 lt=

602 lt=

580 lt=758 gt 758 lt=

742 gt 742 lt=

630 lt= 630 gt

668 gt668 lt= 668 gt668 lt=

680 gt680 lt=

728 lt=728 gt

660 lt= 660 gt 500 gt

602 gt

O

O

OOO

O O O O

O



119864 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (7)


119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (9)

119864 (119904119906119887 119890119903119903119900119903 119888119900119906119899119905) minus V119886119903 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905)gt 119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) (10)







s2_mem

s2_cpu

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 0 2 0 2

1

1

746 gt746 lt=

500 lt=

580 lt=

0 2 1 01 0 2 0 22

580 gt

1 0 2680 gt

620 gt 620 lt=

602 lt=

680 lt=758 gt 758 lt=

500 gt

602 gt

0

O O O O

O







Xen Server2X

eth0

SW8

SW2SW7GRE1

uo0 network

100080

101010130

eth0

SW3

SW9

GRE

uo1 network 1000160

101010116

eth0

SW4

SW5

SW1GRE

core network

10004

10101012610007GRE2

SW6

10003

1000510001

streaming server

streaming server

broadcaster

watcher1000102


Xen Server1

Xen Server1

OpenVirtex



IT operator

Web GUI

Slice manager

Tunnel

Tunnel


Hypervisor

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0

101010116 101010126

Tunnel

s2

gre

s1

gre


15155

16165

17175

0 5 10 15 20 25 30 35 40 45 50 55 60

band

wid

th (G

bps)

time (sec)

GREVxLAN



Hypervisor 1

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0101010116 101010126

Tunnel

s2gre

s1

gre

Hypervisor 2


860880900920940960

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60band

wid

th (M

bps)

time (sec)

VxLAN








(11)



eth1eth0

SW1

SW4

SW2

SW3

BGGRE

eth1eth0

SW1

SW4

SW2

SW3

BGGRE


1000100 101010170 1000200 10101046

Tunnel Slice

Non-Tunnel Slice

192003192001

101010174

SDN Controller

Openvirtex

Xenserver

TunnelController


10001

10003


0102030405060708090

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60

Band

wid

th (M

byte

s)

time (sec)

VxLAN







0

5000

10000

15000

20000

25000

30000

35000

00102030405060708

num

ber o

f req

uest

s

Ratio

of r

edir

ectio

n

DTBFR

MLF

DTBFR ratio

MLF ratio



5 Conclusions



Data Availability




Acknowledgments


References
















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia





Micro Operator








(O)












Core network

BG

BGOVS

OVS

OVSOVS

BG

Micro Operator

Datacenter

BGBG

3G

4GLTE

OpenVirtex

tunnel

tunnel tunnel

tunnel

tunnel

NFVI

NFVI


Slice 3

Slice 2

Slice 1



tunnel











119864119899119905119903119900119901119910 (119904) = 119888sum119894=1

119875119894 log2 119875119894 (1)



119866119886119894119899 (119878119860) = 119864119899119905119903119900119901119910 (119878)minus Vsum119895=1

1003816100381610038161003816100381611987811989510038161003816100381610038161003816|119878| 119864119899119905119903119900119901119910 (119878119895) (2)

119878119901119897119894119905119868119899119891119900 (119878 119860) = minus 119888sum119894=1

10038161003816100381610038161199041198941003816100381610038161003816|119878| log2 (10038161003816100381610038161198781198941003816100381610038161003816|119878| ) (3)

119866119886119894119899119877119886119905119894119900 (119878 119860) = 119866119886119894119899 (119878 119860)119878119901119897119894119905119890119868119899119891119900 (119878 119860) (4)







Start Training data





Unselected_labels



maximum Gain ratio




error_leaf

NO

YES

Pruning

YES


Pruned Tree created

End

NO

YES

NO


error_leaf






119864 (119897119890119886119891 119890119903119903119900119903 119903119886119905119890) = (119864 + 05)119873 (5)

119864 (119904119906119887119905119903119890119890 119890119903119903119900119903 119903119886119905119890) = (119864 + 05 lowast 119897119890119886119891)119873 (6)





s2_mem

s2_cpu s2_cpu

s1_mem

S1_cpu

s0_cpu

s0_mem s0_mem

1 0

0

1 0 2

S1_mem

0

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 2 0 0 2

0

1 0 1 0

1

11 0

746 gt 746 lt=

500 lt=

580 lt=

1 0 1 0 1 0

0 2 0 21 0 1 0 22

1 0 2

1 0 580 gt

1 0 2580 gt

620 gt 620 lt=

602 lt=

580 lt=758 gt 758 lt=

742 gt 742 lt=

630 lt= 630 gt

668 gt668 lt= 668 gt668 lt=

680 gt680 lt=

728 lt=728 gt

660 lt= 660 gt 500 gt

602 gt

O

O

OOO

O O O O

O



119864 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (7)


119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (9)

119864 (119904119906119887 119890119903119903119900119903 119888119900119906119899119905) minus V119886119903 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905)gt 119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) (10)







s2_mem

s2_cpu

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 0 2 0 2

1

1

746 gt746 lt=

500 lt=

580 lt=

0 2 1 01 0 2 0 22

580 gt

1 0 2680 gt

620 gt 620 lt=

602 lt=

680 lt=758 gt 758 lt=

500 gt

602 gt

0

O O O O

O







Xen Server2X

eth0

SW8

SW2SW7GRE1

uo0 network

100080

101010130

eth0

SW3

SW9

GRE

uo1 network 1000160

101010116

eth0

SW4

SW5

SW1GRE

core network

10004

10101012610007GRE2

SW6

10003

1000510001

streaming server

streaming server

broadcaster

watcher1000102


Xen Server1

Xen Server1

OpenVirtex



IT operator

Web GUI

Slice manager

Tunnel

Tunnel


Hypervisor

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0

101010116 101010126

Tunnel

s2

gre

s1

gre


15155

16165

17175

0 5 10 15 20 25 30 35 40 45 50 55 60

band

wid

th (G

bps)

time (sec)

GREVxLAN



Hypervisor 1

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0101010116 101010126

Tunnel

s2gre

s1

gre

Hypervisor 2


860880900920940960

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60band

wid

th (M

bps)

time (sec)

VxLAN








(11)



eth1eth0

SW1

SW4

SW2

SW3

BGGRE

eth1eth0

SW1

SW4

SW2

SW3

BGGRE


1000100 101010170 1000200 10101046

Tunnel Slice

Non-Tunnel Slice

192003192001

101010174

SDN Controller

Openvirtex

Xenserver

TunnelController


10001

10003


0102030405060708090

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60

Band

wid

th (M

byte

s)

time (sec)

VxLAN







0

5000

10000

15000

20000

25000

30000

35000

00102030405060708

num

ber o

f req

uest

s

Ratio

of r

edir

ectio

n

DTBFR

MLF

DTBFR ratio

MLF ratio



5 Conclusions



Data Availability




Acknowledgments


References
















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



Core network

BG

BGOVS

OVS

OVSOVS

BG

Micro Operator

Datacenter

BGBG

3G

4GLTE

OpenVirtex

tunnel

tunnel tunnel

tunnel

tunnel

NFVI

NFVI


Slice 3

Slice 2

Slice 1



tunnel











119864119899119905119903119900119901119910 (119904) = 119888sum119894=1

119875119894 log2 119875119894 (1)



119866119886119894119899 (119878119860) = 119864119899119905119903119900119901119910 (119878)minus Vsum119895=1

1003816100381610038161003816100381611987811989510038161003816100381610038161003816|119878| 119864119899119905119903119900119901119910 (119878119895) (2)

119878119901119897119894119905119868119899119891119900 (119878 119860) = minus 119888sum119894=1

10038161003816100381610038161199041198941003816100381610038161003816|119878| log2 (10038161003816100381610038161198781198941003816100381610038161003816|119878| ) (3)

119866119886119894119899119877119886119905119894119900 (119878 119860) = 119866119886119894119899 (119878 119860)119878119901119897119894119905119890119868119899119891119900 (119878 119860) (4)







Start Training data





Unselected_labels



maximum Gain ratio




error_leaf

NO

YES

Pruning

YES


Pruned Tree created

End

NO

YES

NO


error_leaf






119864 (119897119890119886119891 119890119903119903119900119903 119903119886119905119890) = (119864 + 05)119873 (5)

119864 (119904119906119887119905119903119890119890 119890119903119903119900119903 119903119886119905119890) = (119864 + 05 lowast 119897119890119886119891)119873 (6)





s2_mem

s2_cpu s2_cpu

s1_mem

S1_cpu

s0_cpu

s0_mem s0_mem

1 0

0

1 0 2

S1_mem

0

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 2 0 0 2

0

1 0 1 0

1

11 0

746 gt 746 lt=

500 lt=

580 lt=

1 0 1 0 1 0

0 2 0 21 0 1 0 22

1 0 2

1 0 580 gt

1 0 2580 gt

620 gt 620 lt=

602 lt=

580 lt=758 gt 758 lt=

742 gt 742 lt=

630 lt= 630 gt

668 gt668 lt= 668 gt668 lt=

680 gt680 lt=

728 lt=728 gt

660 lt= 660 gt 500 gt

602 gt

O

O

OOO

O O O O

O



119864 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (7)


119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (9)

119864 (119904119906119887 119890119903119903119900119903 119888119900119906119899119905) minus V119886119903 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905)gt 119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) (10)







s2_mem

s2_cpu

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 0 2 0 2

1

1

746 gt746 lt=

500 lt=

580 lt=

0 2 1 01 0 2 0 22

580 gt

1 0 2680 gt

620 gt 620 lt=

602 lt=

680 lt=758 gt 758 lt=

500 gt

602 gt

0

O O O O

O







Xen Server2X

eth0

SW8

SW2SW7GRE1

uo0 network

100080

101010130

eth0

SW3

SW9

GRE

uo1 network 1000160

101010116

eth0

SW4

SW5

SW1GRE

core network

10004

10101012610007GRE2

SW6

10003

1000510001

streaming server

streaming server

broadcaster

watcher1000102


Xen Server1

Xen Server1

OpenVirtex



IT operator

Web GUI

Slice manager

Tunnel

Tunnel


Hypervisor

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0

101010116 101010126

Tunnel

s2

gre

s1

gre


15155

16165

17175

0 5 10 15 20 25 30 35 40 45 50 55 60

band

wid

th (G

bps)

time (sec)

GREVxLAN



Hypervisor 1

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0101010116 101010126

Tunnel

s2gre

s1

gre

Hypervisor 2


860880900920940960

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60band

wid

th (M

bps)

time (sec)

VxLAN








(11)



eth1eth0

SW1

SW4

SW2

SW3

BGGRE

eth1eth0

SW1

SW4

SW2

SW3

BGGRE


1000100 101010170 1000200 10101046

Tunnel Slice

Non-Tunnel Slice

192003192001

101010174

SDN Controller

Openvirtex

Xenserver

TunnelController


10001

10003


0102030405060708090

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60

Band

wid

th (M

byte

s)

time (sec)

VxLAN







0

5000

10000

15000

20000

25000

30000

35000

00102030405060708

num

ber o

f req

uest

s

Ratio

of r

edir

ectio

n

DTBFR

MLF

DTBFR ratio

MLF ratio



5 Conclusions



Data Availability




Acknowledgments


References
















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia





119864119899119905119903119900119901119910 (119904) = 119888sum119894=1

119875119894 log2 119875119894 (1)



119866119886119894119899 (119878119860) = 119864119899119905119903119900119901119910 (119878)minus Vsum119895=1

1003816100381610038161003816100381611987811989510038161003816100381610038161003816|119878| 119864119899119905119903119900119901119910 (119878119895) (2)

119878119901119897119894119905119868119899119891119900 (119878 119860) = minus 119888sum119894=1

10038161003816100381610038161199041198941003816100381610038161003816|119878| log2 (10038161003816100381610038161198781198941003816100381610038161003816|119878| ) (3)

119866119886119894119899119877119886119905119894119900 (119878 119860) = 119866119886119894119899 (119878 119860)119878119901119897119894119905119890119868119899119891119900 (119878 119860) (4)







Start Training data





Unselected_labels



maximum Gain ratio




error_leaf

NO

YES

Pruning

YES


Pruned Tree created

End

NO

YES

NO


error_leaf






119864 (119897119890119886119891 119890119903119903119900119903 119903119886119905119890) = (119864 + 05)119873 (5)

119864 (119904119906119887119905119903119890119890 119890119903119903119900119903 119903119886119905119890) = (119864 + 05 lowast 119897119890119886119891)119873 (6)





s2_mem

s2_cpu s2_cpu

s1_mem

S1_cpu

s0_cpu

s0_mem s0_mem

1 0

0

1 0 2

S1_mem

0

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 2 0 0 2

0

1 0 1 0

1

11 0

746 gt 746 lt=

500 lt=

580 lt=

1 0 1 0 1 0

0 2 0 21 0 1 0 22

1 0 2

1 0 580 gt

1 0 2580 gt

620 gt 620 lt=

602 lt=

580 lt=758 gt 758 lt=

742 gt 742 lt=

630 lt= 630 gt

668 gt668 lt= 668 gt668 lt=

680 gt680 lt=

728 lt=728 gt

660 lt= 660 gt 500 gt

602 gt

O

O

OOO

O O O O

O



119864 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (7)


119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (9)

119864 (119904119906119887 119890119903119903119900119903 119888119900119906119899119905) minus V119886119903 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905)gt 119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) (10)







s2_mem

s2_cpu

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 0 2 0 2

1

1

746 gt746 lt=

500 lt=

580 lt=

0 2 1 01 0 2 0 22

580 gt

1 0 2680 gt

620 gt 620 lt=

602 lt=

680 lt=758 gt 758 lt=

500 gt

602 gt

0

O O O O

O







Xen Server2X

eth0

SW8

SW2SW7GRE1

uo0 network

100080

101010130

eth0

SW3

SW9

GRE

uo1 network 1000160

101010116

eth0

SW4

SW5

SW1GRE

core network

10004

10101012610007GRE2

SW6

10003

1000510001

streaming server

streaming server

broadcaster

watcher1000102


Xen Server1

Xen Server1

OpenVirtex



IT operator

Web GUI

Slice manager

Tunnel

Tunnel


Hypervisor

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0

101010116 101010126

Tunnel

s2

gre

s1

gre


15155

16165

17175

0 5 10 15 20 25 30 35 40 45 50 55 60

band

wid

th (G

bps)

time (sec)

GREVxLAN



Hypervisor 1

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0101010116 101010126

Tunnel

s2gre

s1

gre

Hypervisor 2


860880900920940960

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60band

wid

th (M

bps)

time (sec)

VxLAN








(11)



eth1eth0

SW1

SW4

SW2

SW3

BGGRE

eth1eth0

SW1

SW4

SW2

SW3

BGGRE


1000100 101010170 1000200 10101046

Tunnel Slice

Non-Tunnel Slice

192003192001

101010174

SDN Controller

Openvirtex

Xenserver

TunnelController


10001

10003


0102030405060708090

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60

Band

wid

th (M

byte

s)

time (sec)

VxLAN







0

5000

10000

15000

20000

25000

30000

35000

00102030405060708

num

ber o

f req

uest

s

Ratio

of r

edir

ectio

n

DTBFR

MLF

DTBFR ratio

MLF ratio



5 Conclusions



Data Availability




Acknowledgments


References
















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia





Start Training data





Unselected_labels



maximum Gain ratio




error_leaf

NO

YES

Pruning

YES


Pruned Tree created

End

NO

YES

NO


error_leaf






119864 (119897119890119886119891 119890119903119903119900119903 119903119886119905119890) = (119864 + 05)119873 (5)

119864 (119904119906119887119905119903119890119890 119890119903119903119900119903 119903119886119905119890) = (119864 + 05 lowast 119897119890119886119891)119873 (6)





s2_mem

s2_cpu s2_cpu

s1_mem

S1_cpu

s0_cpu

s0_mem s0_mem

1 0

0

1 0 2

S1_mem

0

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 2 0 0 2

0

1 0 1 0

1

11 0

746 gt 746 lt=

500 lt=

580 lt=

1 0 1 0 1 0

0 2 0 21 0 1 0 22

1 0 2

1 0 580 gt

1 0 2580 gt

620 gt 620 lt=

602 lt=

580 lt=758 gt 758 lt=

742 gt 742 lt=

630 lt= 630 gt

668 gt668 lt= 668 gt668 lt=

680 gt680 lt=

728 lt=728 gt

660 lt= 660 gt 500 gt

602 gt

O

O

OOO

O O O O

O



119864 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (7)


119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (9)

119864 (119904119906119887 119890119903119903119900119903 119888119900119906119899119905) minus V119886119903 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905)gt 119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) (10)







s2_mem

s2_cpu

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 0 2 0 2

1

1

746 gt746 lt=

500 lt=

580 lt=

0 2 1 01 0 2 0 22

580 gt

1 0 2680 gt

620 gt 620 lt=

602 lt=

680 lt=758 gt 758 lt=

500 gt

602 gt

0

O O O O

O







Xen Server2X

eth0

SW8

SW2SW7GRE1

uo0 network

100080

101010130

eth0

SW3

SW9

GRE

uo1 network 1000160

101010116

eth0

SW4

SW5

SW1GRE

core network

10004

10101012610007GRE2

SW6

10003

1000510001

streaming server

streaming server

broadcaster

watcher1000102


Xen Server1

Xen Server1

OpenVirtex



IT operator

Web GUI

Slice manager

Tunnel

Tunnel


Hypervisor

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0

101010116 101010126

Tunnel

s2

gre

s1

gre


15155

16165

17175

0 5 10 15 20 25 30 35 40 45 50 55 60

band

wid

th (G

bps)

time (sec)

GREVxLAN



Hypervisor 1

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0101010116 101010126

Tunnel

s2gre

s1

gre

Hypervisor 2


860880900920940960

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60band

wid

th (M

bps)

time (sec)

VxLAN








(11)



eth1eth0

SW1

SW4

SW2

SW3

BGGRE

eth1eth0

SW1

SW4

SW2

SW3

BGGRE


1000100 101010170 1000200 10101046

Tunnel Slice

Non-Tunnel Slice

192003192001

101010174

SDN Controller

Openvirtex

Xenserver

TunnelController


10001

10003


0102030405060708090

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60

Band

wid

th (M

byte

s)

time (sec)

VxLAN







0

5000

10000

15000

20000

25000

30000

35000

00102030405060708

num

ber o

f req

uest

s

Ratio

of r

edir

ectio

n

DTBFR

MLF

DTBFR ratio

MLF ratio



5 Conclusions



Data Availability




Acknowledgments


References
















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia





s2_mem

s2_cpu s2_cpu

s1_mem

S1_cpu

s0_cpu

s0_mem s0_mem

1 0

0

1 0 2

S1_mem

0

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 2 0 0 2

0

1 0 1 0

1

11 0

746 gt 746 lt=

500 lt=

580 lt=

1 0 1 0 1 0

0 2 0 21 0 1 0 22

1 0 2

1 0 580 gt

1 0 2580 gt

620 gt 620 lt=

602 lt=

580 lt=758 gt 758 lt=

742 gt 742 lt=

630 lt= 630 gt

668 gt668 lt= 668 gt668 lt=

680 gt680 lt=

728 lt=728 gt

660 lt= 660 gt 500 gt

602 gt

O

O

OOO

O O O O

O



119864 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (7)


119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) = 119873 lowast 119890 (9)

119864 (119904119906119887 119890119903119903119900119903 119888119900119906119899119905) minus V119886119903 (119904119906119887119905119903119890119890 119890119903119903119900119903 119888119900119906119899119905)gt 119864 (119897119890119886119891 119890119903119903119900119903 119888119900119906119899119905) (10)







s2_mem

s2_cpu

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 0 2 0 2

1

1

746 gt746 lt=

500 lt=

580 lt=

0 2 1 01 0 2 0 22

580 gt

1 0 2680 gt

620 gt 620 lt=

602 lt=

680 lt=758 gt 758 lt=

500 gt

602 gt

0

O O O O

O







Xen Server2X

eth0

SW8

SW2SW7GRE1

uo0 network

100080

101010130

eth0

SW3

SW9

GRE

uo1 network 1000160

101010116

eth0

SW4

SW5

SW1GRE

core network

10004

10101012610007GRE2

SW6

10003

1000510001

streaming server

streaming server

broadcaster

watcher1000102


Xen Server1

Xen Server1

OpenVirtex



IT operator

Web GUI

Slice manager

Tunnel

Tunnel


Hypervisor

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0

101010116 101010126

Tunnel

s2

gre

s1

gre


15155

16165

17175

0 5 10 15 20 25 30 35 40 45 50 55 60

band

wid

th (G

bps)

time (sec)

GREVxLAN



Hypervisor 1

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0101010116 101010126

Tunnel

s2gre

s1

gre

Hypervisor 2


860880900920940960

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60band

wid

th (M

bps)

time (sec)

VxLAN








(11)



eth1eth0

SW1

SW4

SW2

SW3

BGGRE

eth1eth0

SW1

SW4

SW2

SW3

BGGRE


1000100 101010170 1000200 10101046

Tunnel Slice

Non-Tunnel Slice

192003192001

101010174

SDN Controller

Openvirtex

Xenserver

TunnelController


10001

10003


0102030405060708090

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60

Band

wid

th (M

byte

s)

time (sec)

VxLAN







0

5000

10000

15000

20000

25000

30000

35000

00102030405060708

num

ber o

f req

uest

s

Ratio

of r

edir

ectio

n

DTBFR

MLF

DTBFR ratio

MLF ratio



5 Conclusions



Data Availability




Acknowledgments


References
















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



s2_mem

s2_cpu

s1_mem

s1_cpu

s0_mem s0_mem

1 2

1 1 2

s1_cpu

1 0 2 0 0 2 0 2

1

1

746 gt746 lt=

500 lt=

580 lt=

0 2 1 01 0 2 0 22

580 gt

1 0 2680 gt

620 gt 620 lt=

602 lt=

680 lt=758 gt 758 lt=

500 gt

602 gt

0

O O O O

O







Xen Server2X

eth0

SW8

SW2SW7GRE1

uo0 network

100080

101010130

eth0

SW3

SW9

GRE

uo1 network 1000160

101010116

eth0

SW4

SW5

SW1GRE

core network

10004

10101012610007GRE2

SW6

10003

1000510001

streaming server

streaming server

broadcaster

watcher1000102


Xen Server1

Xen Server1

OpenVirtex



IT operator

Web GUI

Slice manager

Tunnel

Tunnel


Hypervisor

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0

101010116 101010126

Tunnel

s2

gre

s1

gre


15155

16165

17175

0 5 10 15 20 25 30 35 40 45 50 55 60

band

wid

th (G

bps)

time (sec)

GREVxLAN



Hypervisor 1

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0101010116 101010126

Tunnel

s2gre

s1

gre

Hypervisor 2


860880900920940960

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60band

wid

th (M

bps)

time (sec)

VxLAN








(11)



eth1eth0

SW1

SW4

SW2

SW3

BGGRE

eth1eth0

SW1

SW4

SW2

SW3

BGGRE


1000100 101010170 1000200 10101046

Tunnel Slice

Non-Tunnel Slice

192003192001

101010174

SDN Controller

Openvirtex

Xenserver

TunnelController


10001

10003


0102030405060708090

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60

Band

wid

th (M

byte

s)

time (sec)

VxLAN







0

5000

10000

15000

20000

25000

30000

35000

00102030405060708

num

ber o

f req

uest

s

Ratio

of r

edir

ectio

n

DTBFR

MLF

DTBFR ratio

MLF ratio



5 Conclusions



Data Availability




Acknowledgments


References
















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



Xen Server2X

eth0

SW8

SW2SW7GRE1

uo0 network

100080

101010130

eth0

SW3

SW9

GRE

uo1 network 1000160

101010116

eth0

SW4

SW5

SW1GRE

core network

10004

10101012610007GRE2

SW6

10003

1000510001

streaming server

streaming server

broadcaster

watcher1000102


Xen Server1

Xen Server1

OpenVirtex



IT operator

Web GUI

Slice manager

Tunnel

Tunnel


Hypervisor

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0

101010116 101010126

Tunnel

s2

gre

s1

gre


15155

16165

17175

0 5 10 15 20 25 30 35 40 45 50 55 60

band

wid

th (G

bps)

time (sec)

GREVxLAN



Hypervisor 1

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0101010116 101010126

Tunnel

s2gre

s1

gre

Hypervisor 2


860880900920940960

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60band

wid

th (M

bps)

time (sec)

VxLAN








(11)



eth1eth0

SW1

SW4

SW2

SW3

BGGRE

eth1eth0

SW1

SW4

SW2

SW3

BGGRE


1000100 101010170 1000200 10101046

Tunnel Slice

Non-Tunnel Slice

192003192001

101010174

SDN Controller

Openvirtex

Xenserver

TunnelController


10001

10003


0102030405060708090

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60

Band

wid

th (M

byte

s)

time (sec)

VxLAN







0

5000

10000

15000

20000

25000

30000

35000

00102030405060708

num

ber o

f req

uest

s

Ratio

of r

edir

ectio

n

DTBFR

MLF

DTBFR ratio

MLF ratio



5 Conclusions



Data Availability




Acknowledgments


References
















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



Hypervisor 1

VM1 Controller VM2

10001

Openvirtex

10001

eth0 eth0101010116 101010126

Tunnel

s2gre

s1

gre

Hypervisor 2


860880900920940960

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60band

wid

th (M

bps)

time (sec)

VxLAN








(11)



eth1eth0

SW1

SW4

SW2

SW3

BGGRE

eth1eth0

SW1

SW4

SW2

SW3

BGGRE


1000100 101010170 1000200 10101046

Tunnel Slice

Non-Tunnel Slice

192003192001

101010174

SDN Controller

Openvirtex

Xenserver

TunnelController


10001

10003


0102030405060708090

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60

Band

wid

th (M

byte

s)

time (sec)

VxLAN







0

5000

10000

15000

20000

25000

30000

35000

00102030405060708

num

ber o

f req

uest

s

Ratio

of r

edir

ectio

n

DTBFR

MLF

DTBFR ratio

MLF ratio



5 Conclusions



Data Availability




Acknowledgments


References
















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



eth1eth0

SW1

SW4

SW2

SW3

BGGRE

eth1eth0

SW1

SW4

SW2

SW3

BGGRE


1000100 101010170 1000200 10101046

Tunnel Slice

Non-Tunnel Slice

192003192001

101010174

SDN Controller

Openvirtex

Xenserver

TunnelController


10001

10003


0102030405060708090

GRE

0 5 10 15 20 25 30 35 40 45 50 55 60

Band

wid

th (M

byte

s)

time (sec)

VxLAN







0

5000

10000

15000

20000

25000

30000

35000

00102030405060708

num

ber o

f req

uest

s

Ratio

of r

edir

ectio

n

DTBFR

MLF

DTBFR ratio

MLF ratio



5 Conclusions



Data Availability




Acknowledgments


References
















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia




0

5000

10000

15000

20000

25000

30000

35000

00102030405060708

num

ber o

f req

uest

s

Ratio

of r

edir

ectio

n

DTBFR

MLF

DTBFR ratio

MLF ratio



5 Conclusions



Data Availability




Acknowledgments


References
















































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia











































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia


micro operator design pattern in 5g sdn/nfv network

Documents