Optimization of iot services deployment in cloud fog system

Along with the new network architectures, itenables a new breed of services and applications with tightly Quality of services.The emerging problem is how to efficiently deploy the servic

VIETNAM NATIONAL UNIVERSITY, HANOI UNIVERSITY OF ENGINEERING AND TECHNOLOGY

DANG VAN DO

OPTIMIZATION OF IOT SERVICES DEPLOYMENT

IN CLOUD-FOG SYSTEM

MASTER THESIS Major: Data Communication and Computer Networks

HA NOI - 2019

VIETNAM NATIONAL UNIVERSITY, HANOI UNIVERSITY OF ENGINEERING AND TECHNOLOGY

Dang Van Do

OPTIMIZATION OF IOT SERVICES DEPLOYMENT

IN CLOUD-FOG SYSTEM

MASTER THESIS Major: Data Communication and Computer Networks

Supervisor: Dr Tran Truc Mai

Assoc.Prof Nguyen Kim Khoa

HA NOI - 2019

With the predicted explosion in the number of connected devices, sensors andextremely large amount of data generated need to be analyzed, the current cloudparadigms, which tend to me concentrate computing and storage resources in afew large data centers, will inevitably lead to excessive network load, end-to-endservice latency, and overall power consumption This leads to the creation of newnetwork architectures that extend computing and storage capabilities to the edge

of the network, close to end-users Along with the new network architectures, itenables a new breed of services and applications with tightly Quality of services.The emerging problem is how to efficiently deploy the services to the systemthat satisfies service resource requirements and QoS constraints while maximizingresource utilization

In this thesis, we investigate the problem of IoT services deployment in Fog system to provide IoT services with minimal resource usage cost We for-mulate the problem using a Mixed-Integer Linear Programming model taking intoaccount the characteristics of computing and transmission resources in Cloud-Fogsystem as well as the IoT services specific requirements Our solution provides

Cloud-a multi-lCloud-ayer mCloud-apping mechCloud-anism thCloud-at efficiently deploys IoT services to the Cloud-propriate virtual network in physical infrastructure Unfortunately, our proposedmodel is unable to solve in polynomial time due to it is NP-hard We proposegreedy-based algorithms for solving the problem which tries to solve each phase

ap-of the deployment process sequentially We illustrate the utility ap-of our solutionsover a motivating example where we compare the efficiency of our solutions withthe existing solutions for a traffic monitoring service The experimental resultsshow that our proposed solution outperforms compared to existing solutions interms of energy efficiency

My sincere thanks also go to the Faculty of Information and Technology, versity of Engineering and Technology, Vietnam National University for provid-ing me all the necessary facilities to make this research project easier.

Uni-Finally, I would like to say thanks to my family, my friends who have alwaysbelieved, motivated and supported me throughout the past process to achieve to-day’s results

in my thesis, published or otherwise, are fully acknowledged in accordance withthe standard referencing practices.

Furthermore, to the extent that I have included copyrighted material, I certifythat I have obtained written permission from the copyright owner(s) to includesuch material(s) in my thesis and have included copies of such copyright clear-ances to my appendix

I declare that this thesis has not been submitted for a higher degree to any otherUniversity or Institution

Table of Contents

1.1 Motivation 1

1.2 Problem statement 4

1.3 Research questions 6

1.4 Objectives 7

1.5 Outline 8

2 Literature review 9 2.1 Fog computing and the Internet of Things 9

2.1.1 Definition 9

2.1.2 Reference Architecture 10

2.2 IoT services 14

2.3 Optimal services deployment problem 16

2.3.1 Algorithms 16

2.3.2 Comparison and discussion 20

3 Methodology 22 3.1 System model 22

3.1.1 Network model 22

3.1.2 Service model 23

3.1.3 Virtual layer model 24

3.2 The optimization of IoT services deployment in Cloud-Fog system 24 3.2.1 MILP formulation 24

3.2.2 Deployment model 32

4 Experiment results and discussion 36 4.1 Experiment results 36

4.1.1 Simulation details 36

4.1.2 Simulation scenarios 38

4G Fourth Generation

CPU Central Processing Unit

DC Data Center

Gbps Gigabit per second

IoT Internet of Things

J/bit Joule per bit

Mbps Megabit per secondMCF Multi-commodity FlowMILP Mixed Integer Linear ProgrammingMIPS Millions of Instructions Per Second

NP Non-deterministic Polynomial-time

QoS Quality of Service

TCP Transmission Control Protocol

UDP User Datagram Protocol

List of Figures

1.1 Three-layer Cloud-Fog system paradigm 2

1.2 IoT services in Cloud-Fog system 3

2.1 Fog computing reference architecture [1] 12

3.1 Traffic monitoring service model 23

3.2 Services deployment problem 26

4.1 Smart city infrastructure used in our simulations 37

4.2 Average power consumption of the traffic monitoring service for different amounts of energy consumed by server nodes in idle state 39 4.3 Average power consumption of the traffic monitoring service for different edge node efficiencies 40

List of Tables

2.1 Differences between cloud and fog computing 11

2.2 Comparison of characteristics of related work 21

3.1 Notations 29

4.1 Cloud-Fog system resources 37

as a promising computing paradigm, which can provide elastic resources to theIoT services and applications However, for those applications with low latency,location-awareness, mobility support requirements, the current centralized cloudparadigm, which tends to concentrate computing and storage resources in a fewlarge data centers, will be no longer suitable Recently, Cisco has introduced Fogcomputing as a new paradigm which takes advantage of the extensive resources inthe cloud while being able to expand computing power to the edge of the network,close to end-users [2, 3].

Fig 1.1 illustrates the architecture of a Cloud-Fog system with three cal layers where VNFs are deployed to implement service functions At the edge-most of the network is the device layer which contains numerous sensory nodes.They can be widely distributed at various public infrastructures to monitor theircondition changes over time Each node either collects data (i.e., video, tempera-ture, noise) or performs a certain function (i.e., sprinkle, smart light) Data gener-

hierarchi-Figure 1.1: Three-layer Cloud-Fog system paradigm

ated by IoT devices can be sent to and processed by the VNFs deployed at the fognodes nearby the data sources The fog nodes can be micro clouds, access networkdevices or even user devices, which located in a wide-spread geographical area,together they form the fog layer that lies between the device layer and the cloudlayer Each fog node is connected to and responsible for a group of IoT devices,performing data analysis in a timely manner Thanks to the virtualization technol-ogy, the fog nodes with heterogeneous resources can provide the ability to imple-ment IoT service functions, providing the ability to reduce network load as well

as ensure service QoS constraints including latency and location-awareness Ontop of the architecture is cloud layer consists of a number of powerful servers allo-cated in a few data centers The cloud layer is considered as an unlimited resourcepool providing an ability to host VNFs that process computational-intensive tasks,store a massive amount of data

The IoT services take an important part in the Cloud-Fog system paradigm Ittakes the form of the service layer above the shared physical infrastructure layer

as illustrated in Fig 1.2 A key aspect of increasing service performance andenergy efficiency is the actual deployment of services functions Deployment

decisions need to address the resource requirements of the services, make surethat the services meet its QoS constraints, and reduce the overall running cost ofthe system The allocation of resources in a non-optimal way will result in bothlow-performance of services and an increase in the number of physical servers touse while some of these servers have a very low usage rate.

Figure 1.2: IoT services in Cloud-Fog system

While the basic ideas and theoretical foundations of fog computing have beenestablished, the optimal deployment of IoT services onto the Cloud-Fog environ-ment is still facing many challenges While IoT service functions prefer to behosted at the nearby fog nodes instead of cloud to obtain the low latency and loca-tion tracking, a fog node can only host a limited number of the service functionsdue to its resource capabilities Unlike fog nodes, the cloud is considered as apowerful and unlimited resources pool to deploy IoT service functions However,the cloud is far from IoT device networks, deploying IoT service functions tocloud may cause the increasing of network load and service latency The networkoperator has, therefore, to optimize resource utilization while satisfying strict la-tency constraints when deploying IoT applications While the prior works focus

on application placement problem which tries to map application functions into

physical resources, we go a further step in this work by taking into account amulti-layer mapping where service functions are deployed into virtual resourcesbefore being mapping to physical resources The Cloud-Fog system is composed

of services, virtual resources, and physical resources The virtual layer contains

a network of virtual machines that host IoT service functions and be deployed tothe physical layer The virtual machine has a limited number of flavors that define

a number of parameters in which the virtual machine belongs to We formulate

an optimization problem that performs multi-layer mapping that reduces energycosts and increases resource utilization in both fog and cloud Unfortunately, ouroptimization model belongs to the NP-hard form that challenges any solver to find

an optimal solution in polynomial time We propose a greedy-based algorithm toobtain a near-optimal solution for the problem within an acceptable period of time

In this thesis, we propose a solution for finding optimal IoT services ment in the Cloud-Fog system, where the goal is to find the appropriate virtualmachines for each service function and then place the network of the virtual ma-chines onto Cloud-Fog system infrastructure that minimizes the overall energyconsumption Our contributions can be summarized as follows:

deploy-• We formulate the problem of the combination of IoT services deployment andvirtual machine consolidation in the Cloud-Fog system as a mixed-integer lin-ear program with three layers in the Cloud-Fog system including the servicelayer, virtual layer, and physical infrastructure layer

• We propose a greedy-based solution for solving the problem of optimizeddeployment of IoT services in the Cloud-Fog system in which tries to solveeach phase of the deployment process sequentially

1.2 Problem statement

The Cloud-fog system is considered to be an efficient solution for providingresources to handle newly emerging IoT services with tightly QoS constraints.However, while the basic ideas and theoretical foundations of fog computing havebeen established [1, 2], deploying IoT services onto a Cloud-Fog system is stillfacing many challenges Offloading IoT services to the cloud may result in an

additional network traffic load, increasing unnecessary costs while failing to meetthe latency constraints of delay-sensitive services On the contrary, the computing-intensive functions of IoT services can not be deployed to the devices due to itslimitations on computing power and battery life Furthermore, manual deploy-ment of complex IoT services onto the Cloud-Fog system can be complex, time-consuming and error-prone Therefore, the resource provider has to offer a servicethat ensures optimal automatic deployment of IoT services.

One of the major issues in implementing the service is solving the problem ofoptimal deployment of services functions into physical resources The problemcontains selecting of appropriate virtual machines for service functions and thenassigning existing resources to the network of these virtual machines according tospecific constraints

Typically, an IoT service has its own resource requirements and QoS straints The resource requirements of a service often include computing andtransmission capacity which referred to a collection of processor, memory, storageand bandwidth capacity that guarantees the properly running of the service La-tency is often referred to when considering the quality requirements of a service.The optimal deployment of IoT services is known as a highly complex processthat requires to minimize the mapping costs, ensures the deployed services canmeet its requirements as well as maximize the resource utilization

con-Overall, the following challenges are what we have to face when building theservice that optimizes the deployment of IoT services in Cloud-Fog system:

• Cost - energy consumption

Allocating more resources than required when virtualizing services will incurunnecessary costs, whereas allocating insufficient resources will lead to thepoor performance operation of the services Besides, Cloud-Fog system is aheterogeneity multi-layered system in which each resource has its own pro-cessing, storage, and transmission capabilities as well as energy efficiency,the deployment strategy will determine the operating cost of the services.Our proposed solution has to take into account the capabilities and energyefficiency of computing and transmission resources to minimize the energyconsumption of the system

• QoS constraints

Cost minimization may result in the low-performance of the services Thechallenge here is to provide the IoT services with required QoS constraintswith the optimal energy consumption For example, the cloud may have pow-erful resources with high energy efficiency, however, it is far away from end-user devices while the fog nodes close to end-user have a limitation on com-puting power A good deployment strategy is to deploy computing-intensivefunctions onto the cloud and the delay-intensive functions onto nearby fognodes

Therefore, our solution has to come with a strategy that ensures QoS for IoTservices by taking the location and link delay into consideration

• Resource utilization

The allocation of resources in a non-optimal way will result in an increase

in the number of physical servers to use while some of these servers have avery low usage rate These servers contribute significantly to rising operat-ing costs, low energy efficiency Virtual machines consolidation promises to

be a significant emerging solution to alleviate these problems Basically, theCloud-Fog system is built up of numerous physical servers and each of theseservers can run multiple virtual machines Theoretically, virtual machine con-solidation concentrates target VMs into as small a number of running phys-ical servers as possible according to their resource demands Underutilizedservers should be switched to the sleep mode or switched off so that they con-sume no power [4] A consolidation strategy has to be taken into account tomaximize resource utilization

including resource requirements, latency, and location constraints as well astaking into account the heterogeneous, distributed manner of the Cloud-Fogsystem.

• RQ2: How the problem of optimal deployment of IoT services in Cloud-Fogsystem should be formulated?

The proposed formulation for the problem will use the proposed system modeland define a set of mathematical expressions to represent the system’s con-straints

• RQ3: How can we optimally deploy IoT services onto the Cloud-Fog systemwith given resource constraints in order to meet service requirements andminimize the total energy consumption of the system?

The purpose is to design an algorithm that efficiently allocates compute andnetworking resources to IoT services at minimal cost and maximal resourceutilization while meeting service requirements

1.4 Objectives

Our main objective, in this thesis, is to propose a solution that solves the timal deployment of IoT services in the Cloud-Fog system It is divided intosub-objectives as follows:

op-• O1: Building a mathematical model that represents the IoT services andCloud-Fog system

• O2: Building an optimization model for minimizing the total energy sumption of the system while maintaining the resource requirements and QoSconstraints of IoT services

con-• O3: Design an algorithm that optimizes the IoT services deployment ontoCloud-Fog system running in near real-time The algorithm will collect in-formation about services requirements and substrate resources at a centralizednetwork controller, find the optimal deployment and disseminating the solu-tion to all network nodes

• O4: Carrying our simulations to validate the outperformance of our solutioncompared to existing solutions.

1.5 Outline

This thesis is divided into five chapters organized as follows:

• The first chapter is a general introduction We first present the general contextand motivation of this research Then, the problem statement, the researchquestions and the objectives to be achieved are presented

• The second chapter discusses the technical background and the related work

In this chapter, we provide background knowledge needed to understand sequent materials in the next chapters Then, we present a review of the priorworks that have dealt with the services deployment problem and, based ontheir findings, a synthesis was made to compare the different existing ap-proaches, their limitations and highlight the contributions of this thesis

sub-• The third chapter presents the methodology According to the objectives ofthis thesis, we first present the system modeling and then propose a formu-lation for optimization of IoT services deployment in the Cloud-Fog systemproblem Finally, we design an algorithm to efficiently deploy IoT servicesonto Cloud-Fog system

• The fourth shows the experiment results of our proposed solution

• I conclude my work in chapter fifth and discuss some possible future works

on this problem

Chapter 2

Literature review

2.1 Fog computing and the Internet of Things

This section provides background knowledge of this thesis, including fog puting, IoT and the services deployment problem

of fog computing can reside in multiple layers of a network’s topology and thecloud takes an important part in the architecture

Characteristics of fog computing:

• Low latency and location-awareness: The fog contains of multiple computingnodes located at the edge of the network close to IoT and/or end-user deviceswhich means that Fog Computing supports endpoints with the finest services

at the edge of the network

• Widespread geographical distribution: The fog nodes typically are distributed

in a large geographic area In contrast to traditional centralized cloud, the vices and applications targeted by Fog Computing demand widely distributed

• A large number of network nodes: as a consequence of the wide geo-distribution,

as evidenced in sensor networks in general, and the Smart Grid in particular

• Heterogeneity: Fog nodes come in different form factors, and will be ployed in a wide variety of environments

de-• Support for mobility: It is essential for many Fog applications to communicatedirectly with mobile devices, and therefore support mobility techniques, such

as the LISP protocol1, that decouple host identity from location identity, andrequire a distributed directory system

• Predominant role of wireless access

• Real-time interactions: Important Fog applications involve real-time tions rather than batch processing

interac-• Interoperability and federation: Seamless support of certain services ing is a good example) requires the cooperation of different providers Hence,Fog components must be able to interoperate, and services must be federatedacross domains

(stream-Table 2.1 delineates the differences between Cloud and Fog Computing [5]

2.1.2 Reference Architecture

The Fig 2.1 presents a reference architecture for fog computing proposed

in [1] The bottommost layer contains the end devices (sensors), as well as user devices, edge devices and gateways This layer also includes applicationsthat can be installed in the end devices to enhance their functionality The net-work layer take the responsibble for communication between devices in devicelayer and the between the device layer and the cloud The next layer containscloud resources and services that provide computing power to process IoT func-tions which offloaded to the cloud On top of the cloud layer lays the resourcemanagement services that manage the whole infrastructure and enable quality of

end-1 https://www.lispmob.org/

Table 2.1: Differences between cloud and fog computing

Requirements Cloud computing Fog computing

Access Fixed and wireless Mainly wireless

Service location Within the internet At the edge of the

net-work

No of server nodes Few Very large

Distance (client–server) Multiple hops Only one hop

Support for mobility Limited Supported

Geo distribution Centralised Distributed

volatile/redundant

Figure 2.1: Fog computing reference architecture [1]

service to fog computing applications Finally, the topmost layer the IoT servicesthat leverage fog computing to deliver innovative and intelligent applications toend users [1].

Looking inside the Software-Defined Resource Management layer, it ments many services to optimize the use of cloud and fog resources on behalf ofthe applications The goal of the services is to reduce the cost of using the systemwhile maintaining the performance of IoT services by dynamically using cloudand fog resources

imple-• Flow and task placement: this component keeps track of the state of availablecloud, Fog and network resources (information provided by the Monitoringservice) to identify the best candidates to hold incoming tasks and flows forexecution This component communicates with the Resource Provisioningservice to indicate the current number of flows and tasks, which may triggernew rounds of allocations if deemed too high

• Knowledge Base: This component stores historical information about cation demand and resource demands that can be leveraged by other services

appli-to support their decision-making process

• Performance Prediction: This service utilizes information of the KnowledgeBase service to estimate the performance of available cloud resources Thisinformation is used by the Resource Provisioning service to decide the amount

of resources to be provisioned In times where there is a large number of tasksand flow in use or when performance is not satisfactory

• Raw Data Management: This service has direct access to the data sources andprovides views from the data for other services Sometimes, these views can

be obtained by simple querying (e.g, SQL, or NoSQL REST APIs), whereasother times more complex processing may be required (e.g, MapReduce).Nevertheless, the particular method for generation of the view is abstractedaway from other services

• Monitoring This service keeps track of the performance and status of plications and services and supplies this information to other services as re-quired

ap-• Profiling This service builds resource and applications profiles based on formation obtained from the Knowledge Base and Monitoring services.

in-• Resource Provisioning: This service is responsible for acquiring cloud fogand network resources for hosting the applications This allocation is dy-namic, as requirements of applications, as well as number of hosted appli-cations, changes over time Decision on the number of resources is madewith use of information provided by other services (such as Profiling, Perfor-mance Prediction, and Monitoring) and user requirements on latency as well

as credentials managed by the Security service For example, the componentpushes service functions with low latency requirements to edge of network assoon as free resources are available

• Security: This service supplies authentication, authorization, and phy as required by services and applications

cryptogra-Notice that all the elements and services described are referenced only The actualfog stacks and applications can be built without be use of all the elements, or can

be built with other elements not listed in Fig 2.1 In this thesis, we focus on theResource Provisioning service to provide an optimal deployment stategy for IoTservices in Cloud-Fog system

2.2 IoT services

Potentialities offered by the IoT make possible the development of a huge ber of applications, of which only a very small part is currently available to oursociety Many are the domains and the environments in which new applicationswould likely improve the quality of our lives: at home, while travelling, whensick, at work, when jogging and at the gym, just to cite a few These environmentsare now equipped with objects with only primitive intelligence, most of timeswithout any communication capabilities Giving these objects the possibility tocommunicate with each other and to elaborate the information perceived from thesurroundings imply having different environments where a very wide range ofapplications can be deployed [6]

num-An IoT-Service is a transaction between two parties, the service provider andthe service consumer It causes a prescribed function enabling the interaction withthe physical world by measuring the state of entities or by initiating actions whichwill cause a change to the entities [7].

IoT service requirements

Using Cloud-Fog system for IoT services will bring many benefits such as ing the development and prototyping time with cloud platforms, providing flex-ibility and scalability, pricing savings, etc However, IoT services have specificrequirements that have to be taken into account [1]

eas-• Heterogeneity: Hiding the heterogeneity of IoT devices coming from ent providers to offer a wide range of services is required This can be re-solved by virtualizing IoT gateways for the different vendors and optimizingtheir placement on the cloud This is outside the scope of our work

differ-• Intra-application dependencies: An IoT service, which basically made upfrom many virtual functions, may have feature interaction between thesefunctions inside the same service The performance will be degraded if theseservices are deployed in distant virtual machines

• Increase in traffic demand: Communication between cloud-based functionsand local-based functions incurs additional network traffic overhead Besides,there is a challenge in QoS for different services For example, some stream-ing services implement their own custom protocol like RTP and as networktraffic is mostly TCP and UDP, this can cause a problem

• Timing and location: IoT services are characterized by specific constraintssuch as timing and location constraints First, IoT services affect the realworld and thus the delay of transporting the data from the source to the sinkmust not exceed a certain threshold Second, IoT services interact with a set

of sensors and devices placed at a large geographical area and therefore, someservice functions must remain local So, when being mapped, the distancebetween the local function and the remote function must be considered

In this chapter, we first review existing solutions related to the optimal IoTservices deployment problem Accordingly, we analyze their main advantagesand drawbacks and then highlight the novelty and contributions of our proposedapproach.

2.3 Optimal services deployment problem

One of the major goals of infrastructure providers in both cloud and fog puting is to deploy services to their resources at a minimal cost while satisfies theservice requirements The service deployment process includes selecting appro-priate resource instances to host the set of service functions given their computingrequirements and routing the network flows across these instances that satisfy theservice demands There are major challenges with service performance require-ments, especially with IoT services in the Cloud-Fog system

com-2.3.1 Algorithms

The optimal services deployment problem can be formulated as an integer ear programming problem which is reported to be an NP-hard [8] Finding theexact solutions for this problem is known as practical infeasible Heuristic-based

lin-is an alternative approach that guarantees an acceptable solution for the problem

in a practical manner The execution time of heuristic solutions is low compared tothe exact approach However, the solution of heuristic approach is not guaranteed

to be optimal Meta-heuristic solutions may have better results than heuristic tions as they try to escape from the local optimal to perform an almost acceptablesearch of solution space

solu-Depending on the type of principal approach used to attain the desirable ping, we will divide the application placement’s existing work into the exact ap-proach, heuristic, and meta-heuristic solutions

map-Exact approach

The exact solutions for the optimal service deployment problem can be achievedvia solving the ILP model Several algorithms try to solve the problem such as