Traffic Engineering and Quality of Service in Hybrid Software Defined Networks

Samiullah Mehraban,Rajesh Kumar Yadav

IOT Lab,Department of Computer Science&Engineering,Delhi Technological University,Delhi 110042,India

Abstract: For Future networks,many research projects have proposed different architectures around the globe;Software Defined Network (SDN) architectures,through separating Data and Control Layers,offer a crucial structure for it.With a worldwide view and centralized Control,the SDN network provides flexible and reliable network management that improves network throughput and increases link utilization.In addition,it supports an innovative flow scheduling system to help advance Traffic Engineering(TE).For Medium and large-scale networks migrating directly from a legacy network to an SDN Network seems more complicated &even impossible,as there are High potential challenges,including technical,financial,security,shortage of standards,and quality of service degradation challenges.These challenges cause the birth and pave the ground for Hybrid SDN networks,where SDN devices coexist with traditional network devices.This study explores a Hybrid SDN network’s Traffic Engineering and Quality of Services Issues.Quality of service is described by network characteristics such as latency,jitter,loss,bandwidth,and network link utilization,using industry standards and mechanisms in a Hybrid SDN Network.We have organized the related studies in a way that the Quality of Service may gain the most benefit from the concept of Hybrid SDN networks using different algorithms and mechanisms: Deep Reinforcement Learning (DRL),Heuristic algorithm,K path partition algorithm,Genetic algorithm,SOTE algorithm,ROAR method,and Routing Optimization with different optimization mechanisms that help to ensure high-quality performance in a Hybrid SDN Network.

Keywords: DRL;hSDN;QoE;QoS;SDN;TE

I.INTRODUCTION

Professionals and industries are struggling to manage today’s networks.To respond to these difficulties design of Software Defined Network started.The basic idea behind an SDN network is to separate the control and data planes.In recent years,SDN networks have been among the most excellently discussed topics in the telecommunications systems disciplines.It is a novel and critical network design idea that arises with the advancement of Virtualization,Mobility,the Internet of Things,and other technologies.SDN networks are designed to decouple the control and data planes[1].SDN provides reliable and flexible network management.Due to its global perspective and centralized network control,it allows intelligent flow scheduling systems for improving network throughput and link utilization and supports Advanced Traffic Engineering(TD).SDN network is widely used in different scenarios,including the backbone of Google’s Network[2],Microsoft’s public cloud [3],NTT’s edge gateway [4],and optical (IP/WDM) networks [5].SDN networks are used in almost every field of networking,including cloud networks [6,7],data centers[8-10],wide area networks [11-13],and wireless networks [14,15].As per the survey,the SDN market will grow significantly from 8.8 billion USD in 2018 to 28.9 billion USD by 2023 at a Compound Annual Growth Rate(CAGR)of 26%[16].

Deployment of Pure SDN has many limitations and challenges,as the OpenFlow protocol is not developed sufficiently[17],and Commercial controllers and Switches of SDN are not entirely reliable and stable[18].One significant reason that prevents the adaptation of pure SDN is the necessity to implement most standardized network functions in SDN software to get the same functionality as a traditional network provides;this implementation significantly increases the cost of transition to a pure SDN network.For Medium and Large-Scale networks with many traditional devices,migrating directly from a pure conventional network to a Pure SDN network is complicated.There are some limitations and potential challenges,such as(financial challenges,technical difficulties for a seamless migration,lack of standards,security issues,downtime fears,and Quality of Services degradation).Many problems must be solved as a technological challenge,including how resilient,robust,and scalable the centralized Controller can become a single point of failure.As the Internet and networks expand,more than one SDN controller may be required.These issues hold down Full SDN adoption,driving the emergence of a hybrid SDN Network(hSDN)[19],in which SDN and traditional network devices coexist[20].

The literature provides a distinct definition of hybrid SDN networks.A hybrid SDN network is a logical step in migrating from a conventional network to an SDN network in which SDN and traditional devices coexist.It takes advantage of both conventional and SDN networks,the legacy network’s robustness,and the SDN network’s programmability and does not require upgrading all devices to the SDN networks.We can incrementally deploy SDN devices according to network requirements.In a hybrid SDN network,traditional network devices route traffic along the shortest path using distributed routing protocols such as OSPF and SDN devices supporting both OSPF and OpenFlow Protocol.By doing this,traditional network devices can exchange link information with SDN devices and treat each other as legacy devices.The SDN controller can also get information by supporting Link Layer Discovery Protocols(LLDP) and Broadcast Domain Discovery Protocols (BDDP) to get information on the network topology and link status.

When it comes to communication,it indicates how,when,and which devices will communicate with each other to ensure network functions work correctly[18].Many integration levels are described in detail in the paper[21],including integration only at the data layer,integration at the control layer[22],and integration in both layers [23].The Network’s Controller instructs SDN devices to forward packets in hybrid SDN networks.Traditional network devices will continue to employ classic shortest-path routing protocols such as Equal Cost Multipath (ECMP) and OSPF.Once an SDN device introduces to a conventional network,it is necessary to develop an effective routing mechanism for the newly deployed network to increase network performance and link usage.

The fundamental shortcoming of today’s network is the lack of Traffic Engineering and QoS guarantees.Software-defined networks introduced to solve this problem by separating the data and control layers.It provides the dynamicity and flexibility needed to define QoS policies that effectively push them to the data layer devices,to enhance network throughput and link utilization.

In this paper,we survey related research studies on Traffic Engineering and Quality of Services issues in Hybrid IP/SDN networks.We categorize relevant research in a way in which QoS can benefit the most from the Hybrid SDN concept;we studied different schemes,techniques,mechanisms,and algorithms:inter-domain routing mechanisms,Deep Reinforcement Learning (DRL),K path partition,Heuristic algorithm,genetic algorithm,SOTE Algorithm,RL algorithm,ROAR Method,Distributed Algorithm,Routing optimization,Quality of Service Routing,QoS aware Routing,and QoS policy management in a hybrid SDN network.

This work explains the relationships between Traffic Engineering,Quality of Services,and Hybrid Software Defined networks.This survey paper adds scientific value for a better understanding off TE and QoS through Hybrid SDN networks.It may benefit readers interested in studying Traffic Engineering and QoS in Hybrid SDN and SDN networks.Readers will be familiar with the following after reading this survey paper:

• SDN and Hybrid SDN Network architecture

• Review of Traffic Engineering(TE)in the Hybrid SDN network.

• Overview of Quality of Services(QoS)in the hybrid SDN networks

• Other critical and potential challenges for TE and QoS in Hybrid SDN networks

The rest of this article is structured in such a way that first,we review SDN Architectures.Followed by SDN benefits,SDN deployment challenges,Hybrid SDN definition,strategies,models,Traffic Engineering,Quality of Service,Routing optimization,and routing mechanism in a hybrid SDN network,we conclude the article with some closing remarks.

II.SDN ARCHITECTURE

In recent years,Software Defined Networking has been one of the most researched concepts in the telecommunications areas[24].SDN arose in reaction to improving virtualization,mobility,the Internet of Things,and many other technologies.It is a novel and innovative network design idea that separates control layers and data layers and passes network traffic based on commands of a control layer.The network architecture of SDN Networks is divided into two main parts: a logically centralized control layer and a programmable data layer.The split architecture’s control layer contains most network control logic(defined by software programming),simplifying the data layer.As a result,the data layer is controlled exclusively by the installation forwarding decision of the control layer[25,26]SDN simplifies network administration and delivers robust networking programmability.SDN meets and provides continuously changing demands of network end users for network resources,for instance,in cloud computing[27],the Internet of Things[28,29],and NFV (network function virtualization)[30].

SDN network with OpenFlow protocols allows network operators to give flows at a finer granularity than a traditional network with controllers.In classic networks,packets or flows are shared based on a single or few attributes such as destination MAC address,IP prefixes with the most extended destination,or combination of IP address and TCP or UDP port numbers and so on;SDN enables us to treat and manage flows base on more packet header attributes such as Open-Flow Protocol via CDPI(Controller Data Plane interface).

The OpenFlow protocol connects OpenFlow controllers and forwarding planes through communication.It is the first standard communication protocol for SDN that packets traverse over programmable networks,OpenFlow Protocol provides a flow route way,and many versions of OpenFlows are available.The fundamental advantage of the OpenFlow protocol is that it allows multiple manufacturer switches to be set with controllers in the SDN environments.The Controller manages the OpenFlow protocol and instructs switches on managing data packets that arrive.The OpenFlow protocols keep two switch elements: The Flow table and the Secure communication channel,which are both maintained by the protocol and encompass numerous features such as encrypted channels,traffic monitoring,and processing of inbound packets produced by various controllers,among others.

Figure 1 shows the SDN essential Architecture consisting of three different layers.The data or infrastructure layer contains Data forwarding network elements;the Control Layer contains the Controller for controlling the network;and the Application layer,where the applications are situated.

Figure 1.Software-defined network architecture.

2.1 Data Layer

In an SDN network design,the data layer,or the infrastructure layer,is the initial or bottom layer and comprises network components such as access points,routers,virtual switches,physical switches,and so on.A data layer controls data transmission,flow statistics collection,and network monitoring.The Open Flow protocol is the most extensively used CDPI standard for interfacing between control and data layers’ devices[31].A secure channel is an interface that connects remote controllers to the data layer devices,allowing a controller to install and manage devices and switches safely.

2.2 Control Layer

The second layer of the SDN network is the Control layer or Core layer;we can call it the network brain.This layer consists of one or more centralized software-based controllers that set up and administer the data plane and carry network management capabilities.It controls the flow tables and passes the logic table to the data layer.A southbound interface connects the control and data layers[1].Controllers have two major parts: Control logic and functional components.Controllers can run multiple function mechanisms,such as a coordinator,visualizer,etc.

2.3 Application Layer

The last layer of the SDN network is the Application layer,which assists the control layer in configuring and managing networks following the application’s needs.This layer contains one or more network applications according to our network needs for adding new network functionalities,such as Security applications,Routing Protocols,Visualization,Traffic Engineering,etc.The northbound interface connects the application and control layers[32].

2.4 SDN Benefits

The advantages of SDN are apparent: (1) network programmability,which promotes network automation;(2) network administration,which reduces operating costs;and (3) network virtualization.These advantages motivate organizations and network operators to upgrade their conventional networks with SDN-enabled devices and servers.SDN benefits are simply the capacity to regulate flows flexibly and dynamically;once flows pass through an SDN device,the forwarding path might flexibly have controlled;such traffics are referred to as programmable traffic.SDN is utilized for multiple purposes,including traffic engineering [33],flexible routing [34],link failure recovery [35],power savings [36],and safe updates[37].SDN has been extensively researched and used for campus networks [17],data center networks[38],wide area networks [2],and internet exchange points[39].To avoid single-point-of-failure problems,backup controllers can be used[40,41].Popular SDN controllers(ONOS and Open Day Light)often employ three different controllers to offer full service at a single location,and all the controllers interact with each other(through Raft[42])to ensure network state consistency[43,44].

2.5 SDN Deployment Challenges

The deployment of Pure SDN has many limitations and challenges.One significant reason that prevents the adaptation of pure SDN is the necessity to implement most standardized network functions in SDN software to get the same functionality as a traditional network provides;this implementation significantly increases the cost of transition to a pure SDN network.For Medium and Large-Scale networks with many standard devices,migrating directly from a pure conventional network to a Pure SDN network is complicated.There are some limitations and potential challenges,such as (financial challenges,technical difficulties for a seamless migration,lack of standards,security issues,downtime fears,and Quality of Services degradation).These issues stall the implementation of pure Software Defined Networks and give rise to the birth of Hybrid Software Defined Networks (hSDN),in which SDN and conventional devices coexist[20].

Many problems need to be solved as technical challenges,including how resilient,robust,and scalable the centralized Controller may be without being the only source of failure.As the Internet and networks expand,more than one SDN controller may be required.A controller in the SDN network is an attractive proposition for an attack in terms of security[45].In the lack of a secure and robust controller,there are possibilities and opportunities for attackers to edit the controller code and modify the underlying network’s behavior.As a result,a strong emphasis on SDN Security is essential to make SDN powerful and more beneficial.Until now,potential vulnerabilities exist across SDN platforms,and there are limited discussions on SDN security in the industry and research community.For instance,the systems for authentication and authorization that allow different companies to use network resources without ensuring proper resource protection have been called into doubt [46].In terms of financial challenges,full deployment of the SDN network requires a significant budget expenditure that most enterprises cannot pay all at once.The new module requires the network administrator to develop,apply regulations,and implement modules in new hardware devices.No well-tested,production-grade techniques are available for medium/large-scale network deployment.

Regarding business challenges,migrating to Full SDN may create service downtime for end customers.As a result,network operators must build trust in implementing a new module over the well-established,time-tested old module.Purchasing and installing SDN devices to replace the traditional network infrastructure would require a significant capital investment and an increased operational load for the internet service provider.ISPs may hesitate to perform a onestep migration from conventional networks to SDN networks.

Traditional Network avoids the above-mentioned implementation costs,stability,and security difficulties.We need to concentrate on building such an efficient and flexible network module that combines both traditional and SDN devices to effectively take advantage and benefit from both and fulfill the limitations and challenges mentioned above.

III.HYBRID SDN NETWORK

A hybrid Software-defined Network is a networking system that combines conventional and SDN network equipment.It interacts to variable degrees to control,alter,configure,and manage network behaviors and optimize the user experience and performance.There is no need to upgrade all conventional network devices to Software Define Networks because both SDN and conventional devices work together in Hybrid SDN Networks to take benefit of both;the programmability of SDN networks and the Robustness of traditional networks.Conventional network devices provide reliable routing,whereas Software Defined Network devices focus on network optimization.In contrast,controllers in Software Defined Network route network traffic from a worldwide perspective.Legacy switches in conventional networks employ distribution algorithms such as IGP to govern traffic routing.However,in the case of a Hybrid Software Define Network design,legacy devices will manage some traffic while Software Define Network controllers will handle the rest.

In hSDN,Network traffic is routed according to the decisions of distributed routing and switching technologies.Traffic Control in Hybrid Software defines networks depending on deployment strategies(service-based,class-base,Island-base,controllerbase,Agent-base,Hal-base,and overlay base).SDN Network will control and transfer some traffic,while others will be maintained with traditional mechanisms.A logically centralized controller judges how to process and forward network traffic.Conventional network devices employ standard protocols such as ECMP (Equal Cost Multipath) and OSPF to forward packets in a Hybrid Software Defined Network.A centralized controller will control the only Software Defined Network device.These advantages make a Hybrid Software Define Network much more attractive and exciting for the networks.

Traditional routers can only support the OSPF protocol.The SDN devices work in a hybrid mode,keeping the SDN protocol(OpenFlow)for communication with the SDN controller and the OSPF protocol for communication with classic or traditional network devices.Traditional routers can only forward traffic via the shortest paths provided in routing tables;therefore,even though flows in SDN devices can dynamically divide for multipath forwarding stated by SDN controller flow entries.As a result,the hybrid SDN architecture combines a centralized network’s flexibility with a traditional network’s Robustness.

When users want to implement a hybrid SDN network,they must keep some considerations and limits in mind.The current legacy network can provide multiple services,and many users may want to keep all these traditional systems;otherwise,giving up these legacy devices would be wasteful.Because SDN is a new technology,the associated Software and hardware may not have been thoroughly tested and thus may be unreliable.Many users with less or limited budget may request a small number of OpenFlow hybrid Switches deployed into the traditional network.Two hybrid deployments methods are there,incremental deployment and replaced deployment for hybrid networks.

While implementing hybrid SDN networks,many issues are considered,including performance,load balancing,speedy failure recovery,and network traffic.One recommended technique is to deploy SDN controllers incrementally with conventional networking to get the benefits of both concepts simultaneously while keeping network performance and reducing the interruptions to networking services from a Traffic Engineering point of view.

3.1 Hybrid SDN Models

An appropriate model of a hybrid SDN network is needed to deploy a hybrid SDN network effectively,correctly,and efficiently.It is required to answer two critical problems: first,how to unite SDN and legacy devices in the presence of a controller,and second,how to get corrected network status so that the Controller of the Network makes forwarding decisions.There are different classifications and approaches for the deployment of hybrid SDN networks.Categories are based on components and architectures on coexistence in the control layer,coexistence in the data layer only,or coexistence in both layers.It depends on the network administrator and network requirement to define a Hybrid SDN deployment technique suited to the network situation to allow the smooth coexistence of SDN and legacy devices.

Figure 2 shows different categories of the Hybrid SDN network based on the coexistence on the control plane only or coexistence in both planes.

Figure 2.Hybrid SDN network models.

There are different strategies for deploying a Hybrid SDN network.The decision on which approach to deploy relies on many aspects,including network type,network services,capital budgets,operational budgets,and performance requirements.

• Agent Base

• Service Base

• Hal Base

• Class Base

• Island Base

• Controller Base

• Overlay Base

These are deployment strategies for Hybrid SDN networks which in Previous surveys[47,21,48]have thoroughly explored and documented all of the deployment above techniques,benefits,limitations,and so on.Readers can read these survey studies for further information on the challenges and benefits of various deployment strategies.

IV.TRAFFIC ENGINEERING IN HYBRID SDN

Traffic volume has grown dramatically during the last few decades,much like the Internet’s fast expansion.Traffic Engineering (TE) has received a lot of interest in both business and academics as it is an effective method to minimize network congestion and increase network performance.Traffic engineering is a powerful technique that aims to measure and analyze Realtime network traffic and design routing mechanism for improving traffic routing and balancing network flows.Traffic Engineering is a method that enhances network performance by studying data transmission behavior over communication networks.

The primary purpose of Traffic Engineering(TE)is to optimize network performance to enhance network reliability[49].It may accomplish by making the network fault-resilient,eliminating network congestion by balancing the links’load,and so on.Offering TE in hybrid SDN is challenging because traditional devices are not controlled fully through an SDN controller.The restrictions enforced by the SDN controller to enable TE are only applicable to SDN devices,generally leaving legacy device behavior unchanged.The most prevalent traffic engineering strategies in classical networks are MPLS [50] which employs labels,and GMPLS [51],which expands MPLS to manage new switching technologies and interfaces.However,in an SDN network centrally controlled Controller communicates with forwarding components to maintain the worldwide view,traffic demand,network topology,and link state information.Direct routing paths make the SDN network unique,and a leading candidate for many Traffic Engineering solutions;different methods for SDN networks have been proposed[52-54].

With the evolution of SDN and the division of control and data layers,logical centralization of the control layer,SDN has a practical and flexible way of optimizing routing.As the Controller Has Control of flow routing in the control plane,traffic may be dynamically separated and sent to multiple pathways using SDN switches.Compared to the ECMP used in the standard OSPF protocol,this traffic routing gives a more flexible approach to optimizing routing.

In a pure SDN network,programmability and global centralized network activity control are efficient aspects.Because the flow splitting ratio on SDN switches is variable,using specific complete TE methods in a hybrid SDN network is feasible.To improve the profitability of centralized control,researchers must select a suitable migration sequence,which includes numerous optimization strategies.These approaches may be applied not just to hybrid network deployment but also to other network optimization techniques.For example,migration solutions may expand to include: (1)-identifying a suitable placement for the middlebox in the legacy network to get perfect all-right network control[55],(2)-for energy requirements,wireless networks must address the issue of AP location.[56].(3)-Enhance data center utilization by resolving the VM replacement problem[57].

This part will review the Traffic Engineering technique suggested for the Hybrid SDN network.Load balancing is a Traffic Engineering strategy that aims to optimize the allocation of traffic loads across different resources based on a given performance requirement.In the hybrid SDN network,the TE goal of maximizing link utilization may be met by traffic flow load balance to make network congestion accessible since load balancing seeks to determine the average link rate usage for all the network links.

In traditional networks,the OSPF protocol is one of the most used IGP protocols;each link is given a cost or weight and uses the Shortest path with minimum cost forwarding.Traffic is routed between the source and destination address and distributed equally when multiple shortest paths are encountered using Equal Cost Multipath forwarding.Because all nodes in a hybrid SDN network do not support OpenFlow,flow abstraction,complete packet matching,or packet filtering techniques,TE applies a variety of optimization tactics to get the optimum network performance characteristics,etc.In the hybrid SDN network,it is feasible to use particular extensive Traffic Engineering methods because the flow splitting ratio on the SDN device is arbitrary.To optimize the profitability of centralized control,researchers must select an appropriate migration sequence comprising numerous optimization strategies.Traffic Engineering is more challenging in a Hybrid SDN network;solutions based on static routes or ACLS(access control list)provide restricted functionality[58].

Traffic measurement entails gathering,measuring,and monitoring network status data.Traffic measurement is difficult in the hybrid SDN network because only a small percentage of devices are SDN enabled.Traffic management is the study of controlling and arranging network traffic based on network status data provided by traffic measurement.

In Hybrid SDN networks,measuring all flows is unnecessary and too costly.Cheng et al.[59]proposed a method for collecting the load of certain specified connections and estimating the remainder with a 5%margin of error.polverini et al.[60]define and evaluate a valuable criterion for identifying the flows to be monitored that minimize the estimated error dependent on the flow spread parameter.The suggested approach may also assign measurement tasks equitably between network devices while taking forwarding table space into account.

Cheng et al.[59] present a complete traffic monitoring approach that collects real-time load data for all the links in the network.In this strategy,the controller only has to gather the load of a small selection of relevant links before estimating the load of the other links.The suggested method reacts better adapted to dynamic traffic changes than existing methods and can minimize MLU by up to 40%.

To alleviate the effect of network disruption,one intriguing Traffic Engineering approach is to use ECMP to route the majority of traffic flows and SDN to selectively divert a few flows to balance the MLU of the network.However,critical flow rerouting is difficult due to the ample solution space for essential flow selection.Furthermore,building a heuristic solution for this situation is impossible because rule-based heuristics cannot adjust to changes in the traffic matrix and network dynamics.

Zhang et al.[61]explored Hybrid routing to achieve load balancing with supporting several traffic classes.They suggested a hybrid routing system that combines explicit routing with destination-driven routing to provide load balancing with decreased complexity and high scalability improvements.

Guo et al.[62] suggested an RL (Reinforcement Learning)-based strategy for learning trafficsplitting agents to manage constantly changing traffic to achieve and balance links load in a Hybrid SDN network.They provided an acceptable simulation environment to be built when traffic splitting policies are used to eliminate routing loops.They assess their proposed work on real traffic with different topologies,demonstrating that this strategy provides equivalent network performance and quickly creates good routing strategies.

Wang et al.[63] investigate the efficiency of Traffic Engineering in Hybrid SDN networks depending on traffic distribution and forwarding graphs structure.They defined a coherent forwarding graph and built forwarding graphs with high throughput potential while preserving consistency.Their finding reveals that their forwarding graph creation method outperforms other graph development methods regarding throughput and load balancing.

Yash Sinha et al.[64]created a hybrid SDN network paradigm that used SDN and MPLS-based capabilities and was empirically tested using a prototype implementation.Deployment of this Hybrid network model provides proof of the coexistence of SDN and MPLSbased network devices in the network.This architecture provides a better alternative for a more seamless transition from conventional to SDN networks.

Figure 3 depicts the tunnel splicing method with penultimate Hop Popping.The link translator configures forwarding rules for each router along the way from the penultimate router n1 to router 1.The tunnel splicing procedure entails either issuing a label and implementing MPLS forwarding rules or installing OpenFlow entries,depending on the routers involved.

Figure 3.Process of tunnel establishment.

Tu et al.[65] present a tunnel splicing technique for MPLS and OpenFlow router-based heterogeneous networks.The suggested paradigm was created on a Linux system and tested on test networks.The result of emulation is also deployed in several applications and demonstrates its practicality and efficiency.

Nakahado et al.[66] proposed a Hybrid SDN approach,where only edge devices can be updated by SDN devices,while all other devices use classic OSPF routing.The OSPF network’s incoming traffic is distributed vie edge nodes;this method can decrease the congestion ratio in a network.Sharma et al.[67]suggest that conventional network management approaches and controller-based mechanisms in future networks should coexist and operate together to enable progress in deploying SDN devices in traditional networks.

Figure 4 shows a Hybrid SDN network,combining SDN and MPLS-based networks.A mesh of conventional MPLS switches at the network core and SDN devices at the network edges comprise the network topology.The network controller injects rules into the MPLS switch pathways to push and pop the required labels.

Figure 4.Flow rules installation.

We describe relevant Traffic Engineering(TE)work in three areas: IP Traffic Engineering,MPLS Traffic Engineering,and SDN Traffic Engineering.

4.1 IP Traffic Engineering

In paper no [68],Fortz et al.present IP based intradomain Traffic Engineering method to achieve load balance using Equal Cost Multipath;they optimized the weight of OSPF links.They demonstrate that the OSPF weight setting is an N-P complex problem,suggesting a tabu search heuristic algorithm to find the best OSPF weight setting.Their studies show that this heuristic algorithm can discover the best OSPF weight setting.

Experimental results show that this approach outperforms previous technique and algorithms,as traffics is routed based on the shortest path;this is a restriction of route optimization in an IP network.

Sridharan et al.[69] overcome the limitation of equally dividing Traffic over ECMP.For each routing address,a subset of the following hops Multi paths is chosen for this strategy to transfer traffic more efficiently,simulating unequal traffic splitting to get a near-optimum solution.However,this technique necessitates complicated configuration for each route prefix.Xu et al.proposed DEFT [70] and PEFT[71] to obtain optimum Traffic Engineering and proposed a novel link-state routing protocol.As they provide ideal performance in their strategy,traffic is not restricted to the shortest path but may route to the longest path with an increase in the penalty;the limitation of this strategy is that multiple commodity devices do not support it.

Using objective genetic functions,Xu et al.[72]present optimum intra-domain traffic engineering as a multi-commodity flow maximization issue.Adding a second weight to each link expands the OSPF routing system,allowing for flexible Traffic splitting among all shortest paths.In general,to attain optimal routing performance in most existing IP base intradomain TE solutions removing constraints for shortest path routing requires setup or update on present Link state routing protocols,which are sophisticated and errorprone.Conventional switches cannot yet implement these new routing protocols on an extensive scale network.

4.2 MPLS Traffic Engineering

MPLS tunnels are built between source and destination nodes in a Multi-Protocol Label Switching(MPLS) network.Labels added in packet headers allow flows to be transmitted freely among these established tunnels.Elwalid et al.[73] and Kandula et al.[74],for achieving link load optimum routing proposed two online routing protocols for Traffic Engineering (TE) in MPLS networks that divide flows between pre-established label switching paths (LSP).Zhang et al.[75] propose a hybrid MPLS/OSPF network.The Traffic Engineering problem is formulated as a multi-commodities flow problem using linear programming that overcomes the disadvantages of tunnel construction in MPLS Traffic Engineering and limits of shortest route routing in IP networks to enable a suitable Traffic Engineering solution.MPLS Traffic Engineering can dynamically distribute traffic over many tunnels but lacks a global perspective.Furthermore,to forward packets,MPLS TE systems need the formation of MPLS tunnels,in other words,LSPs,which enhance TE complexity and limit scalability.

4.3 SDN Traffic Engineering

Recently in the SDN networks,Traffic Engineering has become a prominent issue.With SDN network development,we can now manage Network flows flexibly and modify their routing in a real-time network.Google [2] and Microsoft [3] have previously built their small-scale SDN enable Inter Data Centers network and are capable of nearly 100%network utilization.Quan et al.discuss a novel SINET-based future internet infrastructure and suggest a SINET-specific solution for crowd cooperation in Software define vehicle networks [76].In [77],Quan et al.describe Software defining adaptive Transmission control systems for choosing different transmission control techniques in a time-varying vehicular environment.In[78],Agarwal et al.control co-flows by suggesting an architecture with near-optimal performance achievement.

Tajiki et al.[79] describe an effective resource reallocation method for Software-defined Data Centers.These efforts are entirely researched based on a pure SDN situation incompatible with Hybrid SDN.

Table 1.IP Traffic Engineering.

[33]This paper is the first research paper to discuss and explain Traffic Engineering issues in hybrid SDN networks.In this paper,Agarwal et al.define TE as linear programming issues.To improve routing and flow balancing in a Hybrid SDN network,they proposed FPTAS(Fully Polynomial Time Approximation System).Likewise,to paper no [33],Hu et al.[80]and Wang et al.[81]proposed a solution for a Hybrid SDN network to maximize Traffic Flows.Hong et al.[82] present a module for the gradual deployment of hybrid SDN networks and suggest numerous heuristic strategies to reduce Maximum Link Utilization from a Traffic Engineering perspective.In paper number[83],Caria et al.deployed SDN at the border routers and divided the OSPF domain into subdomains to provide fine-grained traffic management among subdomains.He et al.[58] investigate TE in two hybrid network modes:hybrid mode and barrier mode.They present a Polynomial Time approximation approach for Traffic Engineering issues with an approximation of(1+w)in the two-hybrid method.

Xu et al.investigate a comprehensive SDN network in which traditional switching combines all devices with SDN switching [84].Chue et al.[35] describe a method for quickly responding to single links breakdown scenarios while avoiding congestion in Hybrid Networks.Vassicchio et al.[85]offer Fibbing,which introduces false nodes to provide more flexible routing by centrally controlling the link state routing protocols.Huin et al.[86]focus on the Traffic Engineering problem to provide a smooth energy-conscious routing system to minimize energy usage.Tajiki et al.present a unique Traffic Engineering concept for an SDN/MPLS Network in a Hybrid SDN network[87].Guo et al.propose SOTE in [88] to improve the SDN node splitting ratio concurrently with the OSPF weight setting in a Hybrid SDN network;in their situation,they jointly optimized the SDN nodes traffic splitting ratio and OSPF weight setting.They optimize traffic splitting for SDN devices to increase flexibility and improve OSPF weight setting to guarantee network primary routing optimization efficiency.

V.QUALITY OF SERVICES IN HYBRID SDN

Suppose we define Quality of service for a network;it is user satisfaction with certain services.It is a composite indicator that assesses user satisfaction Quality of Service(QoS)pertains to any technology that controls data traffic within a network to minimize packet loss,latency,and jitter.We can say that Quality of service means providing specified data transmission in a particular aspect provided by the network.Quality of Service considers bandwidth,packet loss,service level,data correctness,and other factors.For example,QoS guarantees bandwidth if the network offers excellent bandwidth for bandwidth-sensitive data delivery;we do not need QoS if we have a perfect link.QoS indicates low delay if the network delivers low latency transmission service for time-sensitive data delivery.Differentiating application flows is required to achieve QoS since they compete for available network resources.Network resources should be allocated to make the right decision concerning packet forwarding;sometimes,this procedure requires knowledge of cur-rent network states.

The Quality of service provisioning is mainly determined by SLAs(Service Level Agreements)between service providers and end customers.This method does not give finer granular traffic management,yet it performs excellently for best-effort service.However,different application types,such as video conferencing,online gaming,VoIP,and many more,have Flows sensitive to latency,bandwidth,and jitter,necessitating particular handling techniques.There is no established method for imposing constraints on the depth of traffic distinction and setting high-level traffic management strategies.SDN networks divide the network’s data layer and control layer.This division increases network controller control over networks.Controllers can receive a worldwide view of network status.Network applications under the SDN concept are supplied and provide a conceptual network perspective by the network controller,and they are not required to deal with the low-level setup of data plane devices.

Quality of service in a network refers to enough bandwidth,minimum latency,a minimum rate of packet loss,and balancing network link loads.As we know,balancing the link load of the network can cause increasing bandwidth utilization and a decrease in network latency.as a result,balancing the link load requirement of QoS can cover the QoS requirement of latency and bandwidth.

Quality of Service is applied in hard and Soft QoS[89].An accurate and efficient classification method is required to efficiently categorize data flows and meet QoS criteria in SDN and Hybrid SDN networks.

5.1 Internet QoS Models Background

Two primary QoS models exist on the internet: The IntServ (Integrated Service) model (Barden et al.,1994) and DiffServ (differentiated services) model(Blake et al.,1998).We will look at the architecture and key components of both models.

5.2 IntServ Model

To facilitate end-to-end Quality of Service (QoS) for multimedia applications,the IETF suggested the Integrated Services (IntServ) paradigm.Telecommunications and broadband integrated service digital networks(B-ISDN)inspired this concept.It entails using the Resource Reservation Protocol(RSVP)to create a virtual circuit between the source and the destination(Braden et al.,1997).

The RSVP procedure is as follows: the data source assesses its QoS criteria,such as latency,and transmits them to the intended destination via a route message.This information is recorded and forwarded to the receiver by each router along the path from the source to the destination.The destination then sends a RESV PATH message,which travels along the PATH message’s chosen path to conserve resources.Reservations are made in reverse order for each node.Based on its available resources,a node can approve or deny a RESV request.When the reservation process is finished,the source can begin transmitting data.

5.3 The DiffServ Model

The DiffServ or Differentiated services framework was created as an alternative or a supplement to the IntServ framework(Sivavakeesar,2005).This framework achieves scalability in two ways.

• Per-aggregate flow services have replaced perflow services

• shifting complexity away from the network core and toward the network edge.

Furthermore,DiffServ lacks an end-to-end signaling mechanism,making it more appealing for scalability than IntServ.DiffServ defines a set of per-hop packet forwarding techniques that make use of the Type of Service (TOS) field in the IPv4 header.Packets are divided into distinct service classes based on their demands at the edge router.As a result,rather than the flow-by-flow technique used by IntServ,core routers differentiate between packets on a class-by-classbasis.

5.4 QoS Specifications

Generally,Quality of service parameters is classified into two categories,as shown in Figure 5,Metrics and policies.Then Metrics are divided into three categories,Performance,Security,and relative importance.Performance metrics are divided into four groups Timeliness,precision,Accuracy,and Combination.Policies are categorized into two groups,Management and Level of services.

Quality of Service Metrics includes Security,performance,reliability,and cost,which are used to define QoS characteristics.Latency,loss,and data rates are primary performance metrics.

There are rare circumstances when the network load needs to be balanced as traffic grows.However,it is challenging to control network loads.QoS policies are critical in offering management tools for managing network bandwidths.The benefits of QoS policies include security,flexibility,manageability,and performance.

In many network operations,the SDN network concept advantages might boost the Quality of service.As SDN gives a worldwide view of the network,keeping relevant states for the entire path of flows is feasible.Monitoring network statistics at other levels,such as port,per device,per flow,and so on,is also possible.Furthermore,SDN may assist network operators in developing an automated,user-friendly,and robust QoS management framework for their networks through packet scheduling,resource reservation,and queue management.Because network resources are dynamic,Control techniques for QoS provisioning in network applications must be well-defined.

5.5 QoS Monitoring

SDN switches partition the hybrid network into domains,and SDN switches connect each domain to others.To allow QoS-aware forwarding in a hybrid network,various QoS measurements,such as packet loss rate,traffic load,throughput,latency,and so on,must first be collected.These parameters are then combined to provide an accurate representation of network performance.

Traffic Load Monitoring: The OpenFlow protocol may be used to gather load information from the connections connected to SDN devices.For a directed link,we must compute the number of data flows pass-ing through the switch at a given time interval.Then we may compute the number of data flows per unit of time,which shows the link’s traffic load.The flow table on an SDN switch maintains information about flow statistics.Each matching field of a flow entry that corresponds to a defined routing path contains matching information for that path,such as the source/destination IP address,output port,input port,and so on.

Table 3.Comparison of IntServ and DiffServ models.

Figure 6 shows the load monitoring module;the controller of the network sends a Flow State Request message to switch S4 over the OpenFlow control channel at each time interval t and calculates the number of data flows that reach port 2 of S4 and matches with the necessary flow entries e11,e12,...,e2n by the received feedback message.

Figure 6.Hybrid network Traffic load monitoring.

We may acquire traffic load information through the LSA massage on a link not connected to SDN switches.When an SDN device broadcasts LSA messages,an OSPF router receives the message and forwards it to the SDN switches,along with its status.The SDN switch then forwards the received LSA to the controller,who resolves the massage to determine the traffic load on that link.

Packet Loss Rate and Delay: SDN devices partition the network into numerous domains in the Hybrid SDN network,which are often linked by SDN switches [48].The internal domain devices serve as OSPF routers,and each pair of SDN switches may have several OSPF paths.It is difficult to calculate the packet loss rate and latency for each link.The OpenFlow protocol,on the other hand,can be used to determine the packet loss rate and delay of a channel between SDN switches.Despite the fact that traffic inside an OSPF domain does not transit via SDN switches in the hybrid SDN network,a considerable percentage of the traffic will pass between multiple OSPF domains,necessitating its passage through one or more SDN switches.As a result,the packet loss rate and delay may be calculated.

Figure 7 shows the delay measurement in a hybrid SDN network;the delay of each path is separated into two components(round trip delay of data flow and the transmission delay between SDN switches and controller).

Figure 7.Measurement of delay in a hybrid network.

The controller inserts the time t3 into the probe data flow before sending the packet-out massage to S3.S3 transmits the massage to all ports in accordance with the regulations after receiving it.This communication arrives at S4 through an OSPF path and is forwarded as a packet-in message to the controller.The controller gets the massage at t4 and then resolves it to retrieve the sending port of S3,the receiving port of S4,and the time at t3.

Further research has been conducted to optimize QoS in hybrid SDN networks.They proposed various schemes to optimize single QoS metrics or multiple QoS metrics for hybrid SDN networks;they used different techniques to manage traffic to deliver the best possible Quality of service for their created topologies.

Bi et al.[90] suggested a QoS-aware forwarding approach using the SPMC(Single Path Minimum Cost)and K path partition Algorithms to enhance the QoS of industrial applications.Their simulation findings demonstrate that their suggested strategy meets QoS criteria and effectively balances traffic load using OSPF link bandwidth resources in an SDN/OSPF hybrid industrial internet.

Huang et al.[91] present a comparative optimum traffic control approach for hybrid SDN QoS optimization.They tested their strategy with open-source traffic datasets.Their findings indicated that this strategy might significantly enhance network Quality of Service performance metrics like jitter,latency,and link utilization.

Xu et al.[92] present a heuristic approach for incremental deployment with a limited budget.For throughput maximization routing,they use a randomized rounding mechanism and the Depth First search approach.They demonstrated that both algorithms and analysis could outperform theoretical instances.

Their incremental deployment improves performance by roughly 40% over the prior deployment strategy,while their suggested routing algorithm improves throughput by 31%over ECMP in a hybrid network.

Kelkawi et al.[93] try to describe an effective method for the incremental deployment of legacy network switches to SDN enabled,aiming to minimize MLU of the Network.They use a mixture of two meta-heuristic algorithms (Particle Swarm Optimization and Ant Colony Optimization)to find the best sequence.Their simulation findings demonstrated that their suggested technique outperforms previous static migration algorithms,minimizing maximum link utilization by up to 8.5 percent.

Salman et al.[94]provide a technique for ensuring QoS for time-critical applications across hybrid SDN networks.With a 0.1 link usage level,the method obtains a 60%improvement.Their way searches for the shortest path and does not degrade in performance.

Hong et al.[82] designed a system with an SDN controller to answer these questions,which conventional device to upgrade and how SDN and conventional devices interoperated in the hybrid environment.Their result shows that with 20%of devices upgraded to SDN,their system can reduce 32% of Maximum link utilization compared to a traditional network,and compared to a pure SDN network,only an average of 41%flow table capacity is required

Guo et al.[87] proposed the SOTE algorithm of a Heuristic scheme to optimize the traffic splitting proportion of SDN devices and OSPF weight setting.They developed MINLP (Mixed Integer non-Linear programming problem) to investigate the optimization of Traffic Engineering in a Hybrid SDN network.Their finding shows that SOTE can reduce MLU by 24.19%on average and 11.72%in Hybrid SDN compared to Traffic splitting ratio and OSPF weight setting.They also show that with only 10% of SDN switches deployment,we may benefit from a hybrid SDN network.

From the viewpoint of machine learning,Siew et al.[95] address the topic of SDN implementation in traditional networks.For SDN deployment,they introduced the Reinforcement Learning and Markov Decision process approaches.They outlined the generic machine learning process in networking and proposed hybrid SDN implementation in traditional networks using the Markov Decision Process.Model creation and validation can be accomplished using RL techniques such as Q learning or SARSA.

Ren et al.[96],for TE performance,present FRS(Flow Routing and Splitting) algorithm.They simulate their algorithm with different SDN deployment rates,and their findings show that 20%of SDN device deployment FRS methods can achieve minimum link utilization than previous works.

Guo et al.[97]present a heuristic technique,known as a genetic algorithm,to locate migration sequences that will benefit most notably from the viewpoint of traffic engineering.They assess and compare their approach to greedy and static migration strategies.Their findings show that the genetic algorithm outperforms alternative migration sequences,proving that 40% of the adequately deployed router can reap the most benefits.

Jia et al.[19]present a heuristic strategy for deploying SDN devices in hybrid SDN networks.Their plan works in two separate scenarios.1-Increasing network control capability while keeping the upgrade budget unchanged.2-reducing upgrade expenses to gain the finest network control capability.The results demonstrated that this approach could accomplish 95 percent of the controlled flows with only a 10% upgrading cost.This technique can regulate 95 percent of a network’s flow for approximately 10% of the cost of upgrading.

Guo et al.[98] construct and verify the NP-hard problem of TE across Multiple Traffic Matrices.They present a heuristic approach that combines online splitting ratio and offline weight setting optimization to optimize Routing over many TM.On measured traffic datasets,they test their technique with three network topologies.Their findings show that the MLU of the Network may be significantly improved,and their suggested framework outperforms competing approaches.

Hu et al.[80] investigated methods to maximize flow in hybrid SDN networks.They created FPTS(Fully Polynomial Time Approximation Scheme) to solve the issue.When 50 percent of SDN devices are installed,simulation results using real network topologies show that a hybrid SDN network outperforms conventional networks and can attain near-ideal network performance.

Hung Chen et al.[99] describe the problem of hybrid SDN deployment as an optimization problem with resource restrictions.Because this is NP-hard,they apply greedy-based heuristic algorithms and propose the Hybrid Score switch deployment technique.In a hybrid SDN network,Hybrid Score seeks to deliver the highest control capacity given a set number of switches to upgrade.They evaluated their work with real-world enterprise network topology.They found that their Hybrid Score improved the result by 11.28 percent compared to existing strategies.Compared to a fully SDN environment with only 13 percent of SDN switches,their Hybrid Score provides comparable control capacity.

VI.HYBRID SDN OPTIMIZATION STRATEGIES

We will look at prior work in the field of optimization techniques for migrating Networks from pure traditional Networks to Hybrid SDN networks.

Guo et al.[97] present a Traffic Engineering optimization issue and investigate incremental deployment methods for migrating sequences to IP/SDN networks;they estimate the number and location ofrouters to upgrade to minimize Maximum Link Utilization.The author of this research claims that finding an optimal migrating sequence is much more complex and challenging than traveling salesman issues,as MLU is based on previously visited nodes,and the cost of edges is not constant.To solve this problem,they propose greedy and genetic searching algorithms and compare their algorithms with static migration techniques.Their finding demonstrated that the proposed evolutionary algorithm was able to discover a better and an effective migration sequence,and with the migration of 40% of the devices,we can achieve minimized MLU of the Network.

Table 4.Migration and Quality of Service in Hybrid SDN network.

In paper number [20],Kar et al.study migrating from legacy and pure conventional Networks to hybrid SDN networks using a different combination of optimizing hop and path coverage with reducing cost.They proposed a Heuristic algorithm and demonstrated that this problem is an NP-hard problem.They offer MUcPF and MHcPF written in MATLAB program to model route and hop coverage migration sequences.They tested their algorithm against three other algorithms(Accidental coverage,weighted coverage,and Degree coverage algorithms).The simulation results demonstrate that created algorithms provided higher route and hop coverage than current algorithms for the same expense percentage.MUcPF requires just 15% expenditure to obtain 100% route coverage,in contrast to other algorithms’20-30%.

Jia et al.[19]present a heuristic strategy for deploying SDN devices in hybrid SDN networks.Their plan works in two separate scenarios.1-Increasing network control capability while keeping the upgrade budget unchanged.2-reducing upgrade expenses to gain the finest network control capability.The results demonstrated that this approach could accomplish 95 percent of the controlled flows with only a 10% upgrading cost.This technique can regulate 95 percent of a network’s flow for approximately 10% of the cost of upgrading.

Hui Yang et al.[100]proposed a distributed control architecture for SDON (Software defined optical networking)using the block chain technique(BLockCtrl)aiming to detect a single point of failure within the SDON control plane to address the security and efficiency issues faced by SDON.Their simulation result showed that their proposed architecture could significantly decrease the impact of DoS attacks and fault tolerant control in the SDON network.

Yang et al.[101]concentrate on selecting SDN candidate issues to preserve all possible single points of failure in a network which involves selecting a small portion of traditional routers to migrate to SDN controllers.This method intends to gradually update conventional network devices to hybrid SDN networks by limiting and repairing the route length of broken links,decreasing packet delay,and increasing network bandwidth utilization.Performance assessment revealed that the proposed approach reduced the number of SDN switches required while obtaining a relatively short path length repaired.The authors’ second strategy is discovering the repair path that minimizes maximum link utilization following a connection failure.Because of the higher number of repair path candidates evaluated,the author’s technique outperforms the greedy algorithm(2.9%to 15%decrease)with fewer SDN switches.

6.1 Routing Optimization

Huang et al.[90] present DRL-TC,a novel DRLbased architecture,for optimizing network QoS performance in hybrid SDN networks.Based on historical traffic records,they first look for an appropriate SDN migration sequence to maximize total controlled traffic.In contrast to the prior heuristic solution for TE in hybrid SDN networks,DRL-TC focuses on selflearning routing algorithms and adaptive algorithms in various traffic situations.The results reveal that their suggested strategy achieves optimum performance in optimizing maximum link utilization with rising ratios in the deployment of SDN and a significant improvement in lowering network latency and jitter compared to alternative solutions.Routing optimization in conventional distributed networks is primarily concerned with improving OSPF link weight.

As with SDN network development,network operators can manage the network and the traffic flow in a much more flexible way,and in the real-time network,they can change their route selection.Prior research studies focused on the routing optimization of pure SDN networks,which differs from Hybrid SDN networks.Those routing optimizations cannot directly apply to hybrid SDN networks.

Caria et al.[83] implement SDN at the border routers and divide OSPF into sub-domains to provide fine-grain traffic management across subdomains.Xu et al.[102] optimize routing in a particular Hybrid SDN situation;they collaborated to enhance flow routing and gradual deployment in a hybrid SDN network.XU et al.[84] investigate all devices with SDN and classical switching.Vassicchio et al.[85] offer Fibbing,which introduces false nodes to provide more flexible routing by centrally controlling the link state routing protocols.

6.2 Legacy Optimization Mechanisms

The study[48]thoroughly examined the Deployment of Hybrid SDN technologies and discussed traffic optimization methodologies.Agarwal et al.[82]present the FPTAS method to minimize maximum link utilization(MLU).The authors employ a greedy technique to decide SDN site selection based on optimal throughput gain.The analysis assumes a particular traffic matrix that actual traffic will diverge from this traffic matrix.However,no information regarding how to obtain precise traffic patterns is provided.

Guo et al.[88] maximize the splitting ratio of OSPF weight setting and SDN devices.They developed SOTE (SDN/OSPF Traffic Engineering) to reduce the MLU of the Network.However,getting an ideal OSPF link weight configuration in a dynamic network is complex.

[98]investigates Traffic Engineering across multiple TM(Traffic matrices).To improve MLU,they present a heuristic technique.Authors believe multiple traffic matrices can better describe MLU optimization;they employ the K method to cluster traffic matrices and believe numerous TM can better depict dynamic traffic characteristics.

Poularaki et al.[103] look at a broad upgrading approach for hybrid SDN implementation.They are primarily concerned with the impact of upgrading time.Guo et al.[104] investigate a hybrid SDN deployment challenge,including controller and SDN switches.Jia et al.[105] offer a heuristic strategy for hybrid SDN deployment that maximizes network control ability while keeping to upgrade budget constraints.[104,105]mainly cover hybrid SDN deployment strategies.

6.3 AI-Based Optimization Mechanisms

The deployment challenge is different from the traffic control problem.Heuristic methods may be effective in some cases but may not be effectively adaptable to changes in network state [106].Even when utilizing various Traffic mattresses to show network traffic,there may be some scenarios that are not taken into account.Several initiatives have been made;as a result,taking advantage of Artificial Intelligence(AI)techniques such as DRL for Adaptive Routing.Xu et al.[107] describe a unique DRL-based approach to increase end-to-end communication network throughput and discuss DRL’s advantage over previous dynamic control strategies.Yu et al.[108] implement DDPG to improve network link weights in an SDN network based on various incentive goals,and Huang et al.[109]deploy DDPG in the SDN network to optimize the Quality of Experience(QoE).

Zhang et al.[110,111]present a unique Reinforcement learning method for TE in an SDN network.A key strategy of this method is automatically determining which significant flows should be rerouted;it is suited for each traffic matrix but is not ideal for hybrid SDN applications.When deployed,the flows that SDN devices may manage are fixed,and the possibilities for selection are restricted.Sun et al.[112]propose SINET for SDN routing based on DDPG.It picks critic nodes as control nodes.It uses its link weights to build routing pathways.

6.4 Quality of Service(QoS)Routing

Because of providing open control interface by the SDN network,it facilitates a flexible traffic scheduling scheme for the Network;SDN is the appropriate architecture that meets the QoS need of numerous applications,including video and audio.Egilmez et al.[113]propose an Open QoS concept on an Open Flow Controller with support of end-to-end QoS for multimedia distribution,as it is based on Quality of Service routing traffic paths are optimized dynamically to meet needed QoS.They evaluate their open QoS across real networks,and the finding of their experiments reveal that Open QoS guaranteed transmission,and Users report little or no visual artifacts when watching the seamless video.Furthermore,compared to existing QoS systems,guaranteed services are handled in open QoS without affecting any other traffic types in the network.

Yan et al.[114] suggest Quality of Service assurances through SDN.Furthermore,they employ several pathways among source,destination,and queueing methods that provide QoS for various Traffic types.The experimental findings suggest that the proposed HiQoS method may minimize latency while increasing throughput to ensure QoS.

Silver et al.[115]present Network Service Abstraction layer(NSAL)implementation above the management and control layers.The authors present a uniform data model for traditional and SDN devices,allowing for unified management and configuration of both networks to provide Quality Service for time-critical jobs.

Lin et al.[116] present a QAR (QoS aware adaptive routing)scheme for an SDN network with multilayer distributed hierarchical control plan architectures.Mainly it is utilized to decrease signaling latency in large-scale SDN networks through three tiers of controller design.This QAR technique suggests providing adequate time adaptive QoS packet forwarding using an (RL) Reinforcement learning technique and a QoS aware reward function.Simulation finding shows that this QAR technique outperforms the traditional Q learning method,making it realistic for large-scale SDN networks.

Bi et al.[90] investigate the challenge of managing QoS-aware data flows in the context of multiple flows.They present a game theory based on numerous data flow management strategies in the situation of optimum data flow management being an N-P complex issue and capable of minimizing network delay up to(77-98) percent while increasing network throughput by(24-47)percent.

Chienhung et al.[117] present the design of a hybrid SDN network that may use the spanning Tree Protocol to identify the existence of traditional switches to gain a global perspective on the hybrid SDN network.Authors use the Learning Bridge Protocol to let Open Flow switches communicate with legacy switches without requiring any changes to the legacy switches.SDN applications may dynamically discover routing pathways based on pre-defined QoS criteria and current network conditions by using SDN features.

Hui Yang et al.[118] present a novel SUDOI architecture in ubiquitous data center optical interconnection to meet the QoS requirement of extensive user access.The performance of their work is verified on OaaS(Optimization as a Service)testbed for data center services.Their result shows that SUDOI can effectively utilize optical network and application resources in ubiquitous data center optical interconnection without increasing blocking probability.

Alouche et al.[119] present a geographical routing protocol for automotive networks based on hSDN and a clustering technique.It considers three distinct characteristics when deciding which relay is used to transfer the data: 1-contact length among vehicles,2-a load of each vehicle,and 3-a communication fault log integrated with each cluster head.The simulation finding reveals that the suggested approach performs well in terms of throughput,packet top rate,and average routing overhead.

6.5 QoS Aware Routing

Some applications require different QoS metrics,such as packet loss rate,latency,bandwidth,etc.To overcome this issue,we may create SLA for each application to record QoS criteria and assign routing paths accordingly to match QIS (Quality Information System) needs,referred to as QoS-aware Routing.This routing uses GA(Genetic algorithm)to determine the best suitable route,which meets QoS requirements incrementally.Assessment results demonstrated that this algorithm performs well in networks with different QoS needs.

In Data Centers,several big data applications must send vast volumes of data from one place to another;thus,how these application flows are distributed throughout the network to enhance network usage is critical.The authors of Hedera[120]presented an SAbased algorithm to route flows to alternative pathways based on the current network status.This study outperforms the typical ECMP in terms of performance.The inventors of Long [121] suggested a dynamic rerouting method.When the network controller detects traffic load,it utilizes a multi-hop or a single-hop algorithm to shift traffic to other paths.This method can enhance connection utilization compared to the conventional Round Robin Algorithm.

The inventors of LARAC[122]suggested a LARAC QoS-aware routing algorithm that finds near optimum pathways in the Delay Constrained Least Cost issue using a Lagrange relaxation-based aggregated cost technique.Authors in MINA [123] suggested QoS aware flows scheduling technique for IoT devices;this approach uses GA (Genetic algorithm) to repeatedly generate the best-suited path to fulfill QoS criteria.Assessment results demonstrated that this method performs well in networks with different QoS requirements.Only MINA addressed various restrictions in comparable works.MINA’s key challenges include static weight in the cost functions,QoS-aware Routing,and being locked in a local optimum.To address these issues,[117] introduced a QoS aware routing(SAQR) technique based on the simulated annealing(SA)technique to identify the best-suited path that fulfills various QoS standards.

6.6 Interworking and Inter-Domain Routing

The most well-known applications in the early stages of SDN are Google B4 network [2] and Microsoft WAN [3].Google chose SDN to convert the linked WAN Network across data centers,and it is now wholly deployed using Open Flow Network.Before fully transitioning to the SDN network,Google B4 network went to Hybrid deployment in two different phases;the first was completed in 2010,and the second was completed in 2011 when google started to administrator network through the Controller and doubled the capacity of their network.Google began to promote agile development by running both traditional and SDN routing systems in parallel.Hence,SDN routing systems have much higher priority than the classic ones.SDN is deployed incrementally to different Data Centers to allow additional traffic to be moved from the old routing system to the SDN framework.If there is any issue with the SDN,B4 provides the option to turn off the SDN framework and return to conventional routing strategies.During this stage,SDN devices can interfere with legacy routing protocols,and Google deployed consistent routing protocols for SDN applications.In any case,SDN networks and traditional Networks have to communicate with each other even if complete SDN architecture is ultimately deployed.Data centers must exchange data with traditional extermination networks.

An Internet Exchange Point(IXP)is a physical access point of the network in which several ISPs may join their networks and exchange BGP(Border Gateway Protocol)routes.SDX-L3 stands for Software define internet exchange point Layer 3[116].IP prefixes are used in traditional and physical IXP routing and traffic forwarding.SDX enables those to match many header fields;each AS can use remote traffic control.Aside from SDX,the abstraction of virtual switches provides that AS cannot influence or overserve Inter-Domain Routing outside their scope.Some other optimization models updated rules as soon as BGP routes or policies changed.SDX improves the dependability and flexibility of IXPs.

6.7 QoS Policy Management

SLA or Service Level Agreement is essential to create QoS criteria for traffic management of different QoS applications like (video conferencing,video streaming,and online interactive gaming).Each SLA comprises a set of Objectives that are utilized to generate networks level policies translated into devices level primitives(e.g.,packet dropping rate,queue configurations,traffic shaping policies,and forwarding rules).Managing QoS settings in conventional network designs such as MPLS and DiffServ is complex.Conversely,SDN provides a worldwide network view and has robust northbound API capability.It enables network operators to launch services quickly and execute various network regulations.The PolicyCop [124] project intends to provide controllable,adaptable,and vendor-independent administration of QoS rules in SDN/OpenFlow networks using the Northbound SDN API.PolicyCop benefits from OpenFlow features,making it a suitable management framework.It offers on-demand aggregation as well as per-flow control.Policy Cop makes traffic defining easier than MPLS and DiffServ since it does not need network device shutdown.Because of its vendor independence,it is simple to implement in a network and lowers operational overhead.The Policy Cop design has three planes:management,control,and data plane.The data and control planes are standard SDN planes,but the management plane is Policy Cop’s heart.The plane has two aspects: policy validator and policy enforcer;the policy validator identifies and monitors policy breaches,whereas the Enforcer responds to policy violations and enforces network policy regulations.

6.8 QoS Signaling Overhead

Due to the logically centralized nature of the SDN design,all overhead (QoS-related signaling messages)are gathered at the network’s control mechanism(Controller).[125]OpenFlow allows network operators to collect information at various flow levels,such as aggregated and individual flows.Each of these methods of collection has a cost.The per-flow process provides better granularity in terms of QoS-related states.It has a scalability problem.A controller can pool switches in an SDN or OpenFlow network to collect data on active flows.When a flow timeout occurs,it might direct the switch to report flow statistics at a predetermined interval.Another critical consideration is how frequently QoS information should be transmitted from one network device to the Controller.Although getting statistics from the data plan often benefits the Controller in keeping an up-to-date global picture of network conditions,this procedure is a tradeoff between measurement precision speed,signaling overhead,and timeliness and results in the scalability issue of a control plane for a controller[126].For flow statistics,the PayLess[127]technology supports many degrees of flow aggregation through a RESTful API.It uses an adaptive statistics-gathering technique to provide accurate real-time information while incurring minimum network costs.This approach achieves accuracy extremely near that of the constant periodic polling method while reducing massaging overhead by up to 50%compared to the regular polling strategy.

VII.CONCLUSION

The fundamental issue with today’s network is the shortage or even lack of Quality Service guarantees throughout the networks.Researchers and industry pioneers have explored and presented their outcomes&possible solutions to the currently running network design models,while many of these solutions have failed or may not deploy &function properly.Therefore,Researchers have started to use the promising concept of SDN with OpenFlow protocol as they provide a centralized worldwide network perspective with more fine-grained flow management options in the networks.However,many organizations and enterprise networks cannot conveniently afford to migrate directly from traditional networks to an SDN network;by keeping traditional network infrastructure,they might incrementally migrate to a Hybrid SDN network.This paper reviewed Traffic Engineering &Quality of Services Issues in the Hybrid SDN network.With different strategies,we can significantly enhance the overall performance of the network,especially QoS.Incremental deployment improves network performance by around 40% compared to earlier deployment schemes;however,finding an optimal migrating sequence is much more complex and challenging.We have found that with 10-20% of proper SDN switches deployment,we can benefit the most from the concept of Hybrid SDN networks.We believe that the Hybrid SDN network has come a long way;however,some missing areas still need more research and exploration that leads to further improvements.Scalability in the Hybrid SDN network is one of the key problems caused while adding or changing the number of network nodes;for complicated hSDN scenarios,more proven research is needed with scalability studies.Security is another crucial deficiency in the Hybrid SDN network that must be in the essence of it and worthy of being considered as the central part of seamless hybrid SDN networks.In addition to that,Researchers should focus more on privacy issues,new standards,and models of the Hybrid SDN networks.