Three Tier Fog Networks:Enabling IoT/5G for Latency Sensitive Applications

Romana Shahzadi,Ambreen Niaz,Mudassar Ali,＊,Muhammad Naeem,Joel J.P.C.Rodrigues,Farhan Qamar,Syed Muhammad Anwar

1 University Engineering and Technology,Taxila 47050,Pakistan

2 COMSATS University Islamabad,Wah campus,Wah Cantt,47040,Pakistan

3 National Institute of Telecommunications (Inatel),37540-000 Santa Rita do Sapucaí,MG,Brazil

4 Instituto de Telecomunicacções,Portugal

5 University ITMO,49 Kronverksky Pr.,St.Petersburg,197101,Russia

6 University of Fortaleza (Unifor),60811-905 Fortaleza-CE,Brazil

Abstract:Following the progression in Internet of Things (IoT) and 5G communication networks,the traditional cloud computing model have shifted to fog computing.Fog computing provides mobile computing,network control and storage to the network edges to assist latency critical and computation-intensive applications.Moreover,security features are improved in fog paradigm by processing critical data on edge devices instead of data centres outside the control plane of users.However,fog network deployment imposes many challenges including resource allocation,privacy of users,non-availability of programming model and testing software and support for the heterogenous networks.This article highlights these challenges and their potential solutions in detail.This article also discusses threetier fog network architecture,its standardization and benefits in detail.The proposed resource allocation mechanism for three tier fog networks based on swap matching is described.Results show that by practicing the proposed resource allocation mechanism,maximum throughput with reduced latency is achieved.

Keywords:cloud computing; fog networks; matching games; internet of things

I.INTRODUCTION

Evolution of mobile technology from 1G to 5G and applications of Internet of Things (IoT) has contributed to a sharp increase in data storage and processing requirements.Specifically,half of the total mobile data traffic is occupied by mobile video traffic and this percentage keeps on increasing.Though the mobile devices (IoT devices) are getting smarter,they are still incapable of processing large amounts of data by themselves due to limited computing capability and energy resources.This invokes the need for cloud computing model to handle the large amounts of data generated by IoT devices.Cloud computing is a two-tier network architecture,front tier represents mobile network and the back tier encircles cloud devices as shown in figure 1.Data centers or baseband units (BBUs) which stores and process large amount of data form the backbone of cloud computing model.Providing enough energy resources,cost reduction and supporting multiple platforms are major benefits cloud computing can offer.However,low latency and backhaul bandwidth limitations are serious issues of cloud computing.Delay in communication between end IoT devices (EIDs) and cloud proves to be a great barrier,which occurs due to lack of location awareness while providing resources to EIDs.Moreover,cloud computing architecture requires complicated software to interconnect all BBUs using cloud servers.

Edge computing (EC) is an appropriate solution to overcome the challenges of cloud computing,EC provides an intermediate layer (edge layer) between the EIDs and the cloud,reducing computation load at the data centers.Some of the requests which do not require any involvement from the cloud,are transferred to the edge layer devices.This consequently reduces latency in processing the requests and also support mobility.The edge layer can be implemented in different ways depending upon edge devices,network type,the communication protocols used and services offered by the edge layer.Mobile edge computing (MEC),fog computing (FC) and cloudlet computing (CC) as shown in figure 2,are different ways of implementing the edge layer [1].Fog computing offers a computing layer leveraging devices like wireless routers and M2M gateways,known as Fog Nodes (FNs).

Fig.1.Cloud computing model.

Fig.2.Edge computing model.

These FNs are used to compute and store data locally from end devices before forwarding to cloud.MEC offers cloud computing capabilities inside the radio access network (RAN) by deploying intermediate nodes with storage and processing capabilities in the base stations of cellular networks.The cloudlets attempts to provide cloud computing capacity near intelligent devices at the edge of network,which allows end devices to offload computing to the cloudlet devices with resource allocation similar to a data center.Table I differentiates edge computing implementation techniques based on various parameters.

In general,fog computing and networks is a rising platform for computation and storage in an intermediate layer between end user devices and cloud computing data centers.Fog networks deploy FNs close to EIDs and use them to carry out considerable amount of computing tasks.Though in some aspects fog concept is similar to MEC,fog computing is more suitable for IoT applications.Content delivery,mobile big data analytics,and augmented reality are three core scenarios which will benefit from fog the most.Real time applications of fog networking include connected cars,smart grids,smart traffic lights,self-maintaining trains,wireless sensors and actuator networks,decentralized smart building control,IoT and cyber physical system,software defined network,e-health monitoring systems and mobile computing system etc.

Table I.Comparison of Edge Computing Implementation.

Fog networks will help to achieve increased efficiency,low latency,better quality of service (QoS),and location and context awareness.Furthermore,fog network deployment defies many limitations of cloud and eases the deployment of services with low or zero level of tolerance for error,such as industrial and health care applications.However,there are certain challenges which need to be addressed before large scale deployment of fog networks.Security,programmability,supporting heterogeneous networks,scalability,and extremely low latency requirements are some of the key open challenges fog networks are currently facing.

The main contributions of this article are as follows:

·It discusses three-tier fog network architecture,its standardization,and potential benefits.

·It outlines the key challenges in fog networks and their potential solutions.

·It also proposes an efficient FN selection scheme in order to maximize throughput and minimize the latency in fog networks.Rest of this article is organized as follows:Section II begins by elaborating fog network architecture and its role in minimizing many of the limitations of cloud computing model.Section III considers some of the major challenges the fog network can experience,and their proposed solutions in recent literature.Finally,the proposed resource allocation scheme for fog networks is described in section IV and paper is concluded in section V.

II.FOG NETWORK ARCHITECTURE AND ITS BENEFITS

Fog network is a distributed computational paradigm that extends the cloud computing model by transferring data processing and handling closer to EIDs.This will result in fast system response to events by excluding the data round-trip to the cloud.Fog network as shown in figure 3,is a three tier-architecture which encompasses a bottom layer containing end IoT devices (end users),an intermediate fog layer,and a back-end layer consisting of network core layer and cloud data centers.Network core layer acts as a gateway to cloud data center and have dedicated interfaces for communication with fog layer.Fog layer may have multiple FNs which interact with EIDs and process related information.The FNs can be cellular base stations,small cell base stations with enhanced storage and processing capability,and WiFi access points (APs) which can be deployed on fixed locations (such as buildings,road side units) or mobile objects (such as buses,trains,etc.).In current network deployment,the network core layer consists of software defined networking (SDN) nodes which enables extensive governance and precise supervision [2].Packets originating from EIDs have no direct access to cloud,rather they experience another second inspection process that removes all potentially dangerous or problematic contents.As a result,this approach makes network more robust.EIDs are relatively easy to be compromised since they often remain unattended.

Fog network will help to resolves many of IoT-related limitations as follows:

2.1 Bandwidth retention

The phenomenal growth in communication and network technology demands supporting billions of devices.The EIDs generate loads of unnecessary and barren data sets which are given to cloud for processing.This seems to be rather ineffective and useless,as it consumes heavy amount of bandwidth in categorizing processed data as meaningless and null.The collection rate of this data increases constantly which require a certain level of pre-processing at the edge of network before forwarding this raw data to the cloud.In order to minimize bandwidth requirements,traffic costs and necessary cloud storage,trimming of data on the edge is essential [3].Dedicated FNs could resolve this issue of bandwidth requirement through processing and categorizing only the valid data for a fraction of the networking expense.

2.2 Reduced latency and independent processing

For a large number of EIDs,cloud will suffer severe challenges in providing uninterrupted services during irregular connectivity.This connectivity issue will probably be resolved with the advent of 5G,but since redundancy and robustness are required in current deployments,fog networks can be considered a powerful solution.Many industrial and safety critical systems such as patient monitoring platforms,automated production lines and traffic optimization applications,often require endto-end latency of just milliseconds [4].This demand of low latency will be resolved by fog node in 5G [5],[6],though current network deployments are not yet capable to support it.Irrespective of connectivity interruptions,any safety critical system must operate securely and data accumulation should proceed.Once the connection is re-established,data is uploaded to corresponding cloud repository.Dedicated fog node is a perfect solution for such scenarios as it can temporarily store and pre-process data in fog layer,from which mobile network operators (MNOs) may get notifications regarding ill operation and danger.

2.3 Reliable and secure communication

The chance of error increases as more data travel through the network,since bit error rate,packet dropping and data transmission latency are related to actual transmitted data size.IoT applications require uninterrupted and safe services and such an increased error margin cannot be tolerated in safety critical or emergency applications.Fog network reduces the backward propagation of critical data towards the data centre located away from the control of edge user [7].This not only saves the bandwidth but also reduces the chances of attacks due to the reduced path between data centre and edge devices.Thus,local processing done at user end can significantly improve the overall security of network [8].Fog networks permits service consistency and steadiness,and FNs can perform following functions:

·They could possibly act as proxies for delivering security updates and management of sensors.

·They could perform additional operations such as encryption.

·They could detect threats in real time by taking advantage of local information and context.

The major benefit of fog network is none other than aiding networking in the edge,along with all delay-critical services that can be added to the fog layer.

III.CHALLENGES IN FOG NETWORKS AND THEIR POTENTIAL SOLUTIONS

This section identifies and elaborates the existing challenges in fog networks and their potential solutions.

3.1 Computational and radio resource allocation

Evaluating large amounts of data from large number of EIDs requires equally large computing resources.Along with computational resources EIDs require large number of radio resources for high speed connection to FNs.Allocation of the limited computing and radio resources of FNs to all the EIDs to achieve an optimal performance while maintaining quality of experience (QoE) is a key challenge in fog networks.Computation offloading in fog networks subject to radio interference,battery life time of EIDs and delay constraints,is an open optimization problem.

1) Solutions:The authors in [9] provide a matching game (student project allocation game) based solution for joint computing and radio resource allocation problem in fog networks.Authors in [10] combine Stackelberg game and matching theory for the optimization of fog network.A set of EIDs take required data services from a set of FNs which are controlled by service providers.Stackelberg game is used to examine resource allocation problem together with the pricing problem for service providers,followed by many-to-many matching game to study the pairing problem between EIDs and FNs.

3.2 Privacy and security

Fog networks must provide consistent and secure services to EIDs and for this,all devices in a fog network must have a certain level of trust on one another.Authentication alone is inadequate as devices can breakdown anytime and are prone to malicious attacks.Thus,trust should play a two-way role in a fog network,i.e.,FNs that are providing services to EIDs should authenticate their sincerity.Furthermore,EIDs that send the data to FNs should ensure the security of intended node.So,fog network requires a robust trust model to authenticate and ensure secure communication.

1) Solutions:One way to avert the security issues is to shift them from the fog later to back-end layer (cloud data center).In other words,the cloud data center will control the authentication and authorization of EIDs as wells as FNs.Thus,the communication between EIDs and FNs will only involve the computation and storage of data.The authors in [11] proposed identity authentication,data encryption,and data integrity schemes for fog networks with face identification application.These three schemes accompanied by secure hash algorithm-1 (SHA-1) and advanced encryption standard (AES) can deliver integrity,confidentiality,and availability under fog computing in IoT.The authors in [12] presented lightweight privacy preserving data aggregation (LPDA) scheme for fog computing-enhanced IoT,that cannot only fight against incorrect data injection attack,but also aggregate EIDs data into one.

3.3 Supporting heterogeneity

Fog network is heterogeneous due to its location,i.e.,at the edge of network.Therefore,simultaneous data collection from heterogeneous sources and maintain their connection with FNs is the major responsibility of fog network.In IoT scenarios at large scale,handling such heterogeneous network,preserving wireless connectivity and offering services is not easy.

1) Solutions:Deployment of software defined networking (SDN) and virtual private network (VPN) can ease the implementation and management of heterogeneous network.Authors in [13] suggests a new distributed cloud architecture based on blockchain technology with SDN empowered controller FNs at the edge of network.The suggested architecture provides secure,cost-effective,and on-call access to the highest competitive computing infrastructure in an IoT network.The authors in [18] presented SDN as a scalable and feasible solution for providing end to end connectivity between edge computing and heterogeneous networks.The authors in [19] propose a fog computing based gateway to connect heterogeneous sensor networks to IoT.

3.4 Interfacing and programming model

A unified interfacing and programming model is needed for developers to port their applications to fog computing platform.It will allow components to be application aware and permit suitable optimization for different type of applications,forming an application-centric computing.Moreover,it is problematic for developers to orchestrate hierarchical,dynamic and heterogeneous resources to create compatible applications on various platforms.

1) Solutions:Authors in [14] reported a methodology for designing and implementation of a new fog computing paradigm named FogFlow for IoT applications.This programming model not only supports standard interfaces to share and reuse contextual data across services but also permits IoT service developers to program adaptable IoT services easily over cloud and edges.

3.5 Non-availability of testing software for realistic scenarios

With the advent of IoT,many smart sensing devices can be combined with our daily objects.Fog networks compensates for increasing demand of dealing with big data generated,managed and stored by the applications built on sensory networks.Bearing in mind the massive scale and complexity of fog architecture,development and deployment of testing software in such networks could be extremely challenging.For this,pure and efficient simulated environment,capable of representing real scenario is essentially required.

1) Solutions:Authors in [15] addresses the abovementioned issue by giving a pseudo-dynamic testing approach where FNs under test are executed in real environment and a part of the experimental scenario is simulated.

3.6 Supporting vehicular communication

In order to offer an information-rich environment,connected vehicles is becoming a popular trend.Fog computing can possibly overcome the latency problem in vehicular cloud computing (VCC) and fulfil the delay requirement of real-time vehicular services.Integrating fog computing with existing communication infrastructures such as fog-enhanced radio access network (FeRANs) can be a promising solution to effectively support vehicular services [16].Even though the concept of FeRAN has many potential benefits in supporting connected vehicles,numerous challenges have to be considered to make it a reality.Nonetheless,it burdens fog nodes with collection,storage and processing of huge quantity of data generated by vehicles.Efficient handling of mobility of vehicles is among the most critical challenges in accomplishing vehicleto-everything (V2X) service requirements.Additionally,frequent handovers occur in cellular system owing to high speed of vehicles.In FeRAN based V2X networks,for the sake of maintaining service continuity and high performance,along with handovers,services are also migrated from source fog node to target fog node following vehicle's moving trace.As a result,target fog node must have resources available to provide to user during service migration and handover.All these issues need to be considered to make V2X communication a reality.

1) Solutions:A fog resource management in FeRANbased V2X environment is proposed in [16].To support realtime vehicular services,authors in [16] applied fog resource reservation (FRR) and fog resource allocation (FRA) schemes.Authors in [17] have proposed fog network based decentralized public vehicle scheduling system that combines FNs and sensing devices on vehicles.FNs are used to store metadata,which is provided by the sensors on the vehicles.Table II summarizes the technical challenges in fog networks and their potential solutions.

IV.PROPOSED SCHEME FOR FOG NODE SELECTION AND THROUGHPUT MAXIMIZATION

For the fog network of figure 3,we consider that EIDs are attached to FNs on a shared wireless channel while FNs are connected with each other on a dedicated wireless channel operating at frequencies different than those used for communication of FNs with EIDs.Downlink transmission in a fog network is considered,where power is evenly distributed among all EIDs attached to a FN,and as all FNs use same spectrum to communicate with the EIDs,so EIDs will experience interfere from neighboring FNs.A dedicated high speed link is used to connect all FNs to cloud servers.Each one hop link between end IoT device and a FN has different latencylf,which depends bandwidth available,workloadwufand data raterfof FN as follows:wherexuf∈{0,1} is an assignment indicator which indicates that end IoT device connected to a particular FN (xuf=1) or not (xuf=0).We formulated an optimization problem,which aims to maximize the throughput of the fog network by selecting a FN for particular task subject to maximum work load and maximum latency constraints.

Table II.Key Challenges in Fog Networks and Their Potential Solutions:EIDend IoT device,SDNsoftware defined network,VPNvirtual private network.

We propose stable swap matching algorithm,which gives self organizing decentralized solution to the problem of resource allocation [20].FNs and EIDs have individual preferences over one another based on their respective utilities in term of throughput and latency.A matching said to be stable if there does not exist any EIDu2or FNf2,for which FNf1prefers EIDu1over EIDu2,or any EIDu1which prefers FNf2overf1.To achieve a stable network level matching it is necessary that swaps must be carried out if and only if they improve the utility of all the players (i.e.FNs and EIDs {u1,u2,f1,f2}) in matching game.Players involved in matching game keep on changing their preferences depending upon externalities,which results in new EIDFN pairs.

Fig.4.Proposed swap matching algorithm:PLPreference list,FNFog node,EIDEnd IoT device.

The proposed algorithm shown in figure 4,has three phases.In phase-1,initially each EIDu1is attached to a randomly chosen FNf1,(Equivalently,the EID can be initially attached to the nearest FN).EIDu1discovers the near by FNf2.The SINR is determined at each EID,which in turn is used to determine utilities of FNs and EIDs.In phase-2,EIDs and FNs update their preferences and utilities for the current matching.If an EIDu1is not currently attached to its preferred FN (f2),it will send a proposal to FNf2for new matching.When FNf2receives the proposal it will recalculate its utility the new matching and it will accept the matching proposal if and only if the utility off2is increased by accepting the new matching proposal.Otherwise if the proposal is denied,EIDu1will send a proposal to the next FN according to its preference list.Based on current matching,both EIDs and FNs update their corresponding preference lists and utilities at regular intervals.This periodic update ensures that both EIDs and FNs are connected to their corresponding best available option.Phase-3 corresponds to at a stable matching following achieved from the convergence of phase-2 of the proposed algorithm.

4.1 Simulation setup and results

Assume a uniform distribution of number of FNs and EIDs in a fog network.The transmit power of each EID and FN is 13 dBm and 30 dBm respectively.Distance dependent path loss and shadowing affects transmission according to 3GPP specifications.We are assuming uniform distribution of power among EIDs.Each EID needs minimum SINR of 9.5 dB and 174 dBm/Hz is the noise power spectral density.Transmission radius of FN is 50 m.EIDs are varied from a minimum of 20 to a maximum of 100 with an increment of 20.A fullbuffer traffic model for all EIDs is used in our simulations.For comparison,we consider the random allocation of EIDs to FNs which forms a baseline solution for FN selection problem.Figure 5 is the depiction of average sum rate as a function of number of EIDs.Average sum rate increase with the increase in number of EIDs.Proposed scheme is compared with both baseline solution and maximum received signal strength (RSS) based association method.It is clear from figure 5,that proposed scheme for best selection of FNs outperforms baseline solution and maximum received signal strength indicator (RSSI) based association method.

Fig.5.Average sum rate vs number of EIDs.

V.CONCLUSION

Fog is a contemporary network paradigm that aims at providing plenty of computational and storage resources to network edges.This paper presented a brief overview of fog network architecture,its potential benefits alongside key implementation challenges.Fog networking will evolve equally and rapidly with the development in underlying IoT,radio access technologies and innovative edge devices.Furthermore,fog network is a promising solution to many cloud computing limitations but require effective solutions to key challenges like resource allocation,security and mobility.We propose a self organizing decentralized resource allocation mechanism for fog networks.Results show that the proposed scheme outperforms the baseline solution.