Software Deﬁned Network-based Scalable Resource Discovery for Internet of Things

Geo-distributed and heterog eneous Internet of Things (IoT) devices can gener ate hug e amoun t of data. Ine ﬃ cien t manag emen t of IoT-data promotes netw ork congestion and increases computa tional overhead on the data-processing entities. Traditional netw orking architecture, that is lack of functional abstr action and monitoring capabilities, often fails to meet the dynamics of IoT. Softw are Define Netw ork (SDN) can be a viable alterna tive of the traditional netw orking architecture while dealing with IoT. In SDN, manag emen t, monitoring and context sensing of the connected componen ts are simplifie and can be customized. In this paper , SDN-sensed contextual informa tion of di ﬀ eren t componen ts (computa tional entities, netw ork, IoT devices) are combined together to facilita te scalable resource discov ery in IoT. The proposed policy targ ets balanced processing and congestion-less forwarding of IoT-data. Through simula tion studies, it has been demonstr ated that the SDN-based resource discov ery in IoT outperf orms the traditional netw orking based approaches in terms of resource discov ery time and Quality of Service (QoS) satisf action rate.


Introduction
In recen t years, the Internet of Things (IoT) has drawn significa t research interest.Due to rapid enhancemen t in hardw are and comm unica tion technol ogy, it is predicted tha t by 2020, there will be more than 50 billion activ e IoT devices [1].IoT devices are geodistributed, energy constr ained and heterog eneous.The configu ations, applicability and sensing frequency of IoT devices are also div ersified Most of the IoT devices participa te in real-time da ta sensing.As a consequence, the devices can gener ate hug e amoun t of da ta within a minimal time.When a larg e number of IoT devices send da ta sim ul taneousl y tow ards the computa tional entities (e.g.Cloud, Fog nodes, Edg e serv ers), it is more likel y to crea te netw ork cong estion.Besides, random placemen t # Please ensure tha t you use the most up to da te class file available from EAI at http://doc.eai.eu/publications/transactions/latex/ * Corresponding author .Email: m.afrin.ritu@gmail.com of IoT-da ta can increase processing overhead on the computa tional entities.In such scenario, efficiency of the under lying netw ork in managing incoming IoTda ta (da ta processing, da ta forw arding) is very crucial.How ev er, due to lack of functional abstr action and inability in monitoring internal oper ations of the connected componen ts (IoT devices, computa tional entities), the traditional netw orking architecture is not suitable for efficien t IoT-da ta manag emen t.In this case, Softw are Define Netw ork (SDN) can be adopted to overcome the shortcomings of traditional netw orking architecture in respect of IoT [2].
SDN is a very recen t innov ation in netw orking technol ogy tha t oper ates through softw are system in place of specialized and dedica ted hardw are.It offers programmability of netw orking elemen ts by decoupling netw ork control plane and da ta forw arding plane [3].In SDN, there exists a cen tralized entity tha t perceiv es the topol ogy and sta tus of the netw ork.Based on perception, the cen tralized controller entity determines the da ta forw arding rules and notifie the rules to the da ta forw arding entities.Through abstr action of lower lev el netw orking functionalities, SDN can set up, administrate, al ter, and manag e netw ork beha vior dynamicall y.In di fferen t computing par adigms (e.g.Cloud computing, Mobile edg e computing), SDN based sol utions have been expl ored extensiv el y to meet auto-ma ted, on-demand service requests, handle mobility issues, ensure netw ork reliability , etc. [4].SDN-based sol utions promotes virtualiza tion of netw ork, ensures flexibilit in resource utiliza tion, monitors internal oper ations of the connected componen ts, senses contextual inf ormation, minimizes both capital and oper ational expenses.Although, netw orking among the sensors is the fundamen tal factor for IoT [5], SDN-based sol utions for IoT have not been enlightened significa tl y.From the perspectiv e of IoT, SDN-based sol utions can pla y vital roles in resource discov ery and load balancing.
Gener all y, resource discov ery in IoT ref ers to fin appropria te resources for processing IoT-da ta and its associa te routing pa th to forw ard the da ta.In traditional netw orking architecture, the computa tional entities for processing IoT-da ta and the associa te connections are predefine and sta tic.Theref ore, traditional sta tic netw ork architecture can not cope with the increasing number of IoT devices and their uncertain da ta load.As a resul t, QoS degr ada tion in terms of netw ork bandwid th and service deliv ery is widel y observ ed.Taking cognizance of this fact , we investig ate how SDN-based sol utions can facilita te resource discov ery in IoT.The proposed SDN-based sol ution incorpor ates contextual inf orma tion from three di fferen t aspects (computa tional entities, netw ork, IoT devices) while dealing with resource discov ery in IoT to facilita te flexibl da ta processing and cong estion-less da ta forwarding.Besides, the proposed policy ensures dynamic manag emen t of IoT-da ta in SDN tha t can be scalable to certain exten t according to the situa tion.The major contribution of the paper are listed as: • SDN-based sol ution for scalable IoT-resource discov ery to facilita te unin terrupted da ta processing and da ta forw arding.
• Expl ored the applicability of SDN-sensed contextual inf orma tion in managing uncertain load of IoT-da ta.
• Compar ativ e study betw een SDN-based IoT resource discov ery and traditional sta tic netw ork based approach in terms of resource discov ery time and QoS satisf action rate.
In the foll owing section, sev eral rela ted works in this fie d are highlighted (Section.2).In Section. 3 and 4 the system model and SDN-based IoT-resource discov ery are discussed respectiv el y.In Section. 5 perf ormance ev al ua tion is demonstr ated.Section.6 concl udes the paper .

Related Works
Sev eral research works on SDN has already been cond ucted in di fferen t areas of computa tion and netw orking.In [6], authors design a SDN-supported cloud computing environmen t through OpenFLow switches and controllers.They extend the features of OpenFLow controller in order to facilita te load balancing, less energy usag e, and service monitoring.Besides, a queuing model is dev el oped to claim the feasibility of the system.The SDN based sol ution aims at providing QoS satisfie cloud computing services.
In [7] some poten tial architectures of SDN-based Mobile Cloud has been proposed.The authors of the paper focus on iden tifying basic componen ts of SDN-based Mobile Cloud tha t can deal with mobility and uncertain netw ork sta tus.Sev eral frequency selection methods for da ta transmission have also been discussed.The feasibility of the SDN-based sol ution has been highlighted in terms of high packet deliv ery rate and system overhead.
The authors in [8] argued tha t with dense depl oymen t of mobile devices and limited netw ork bandwid th, it becomes di fficul t to assign radio resources for processing service requests.Besides, manag emen t of interf erence and load balancing betw een base sta tions get tough.To overcome these issues, authors propose a softw are define radio access layer named "SoftRAN" .It works as the cen tralized control plane for radio access netw ork.According to the authors, SoftRAN can efficien tl y handle load distribution, manag e interf erence within the netw ork maximize the netw orking throughput.
In respect of scalability in SDN, the authors of [9], claimed tha t SDN scalability is free from inheren t bottleneck.In tha t paper , the scalability of SDN controller has been discussed in details.Besides, the authors investig ate the scalability in SDN in terms of overhead and faul t toler ance.Since SDN red uces netw ork progr amming and manag emen t complexity , SDN enhances the lev el of flexibilit to accommoda te netw ork progr amming and manag emen t at any scale.
The impact of SDN in IoT has also been expl ored in sev eral research works.In [10 ] a softw are define framew ork is proposed tha t sim plifie manag emen t of IoT-driv en process and deals with dynamic challenging aspects of IoT in terms of forw arding, storing and securing sensed IoT-da ta.The framew ork integr ates the softw are define netw ork, softw are define stor ag e, and softw are define security into a single softw are define based control model.
In [11 ] authors represen t a softw are-define IoT system for controlling f ow and mobility in mul ti-netw orks named "UbiFl ow" .UbiFl ow facilita tes controllers entity to be placed distributiv el y so tha t urban-scale SDN can be divided into di fferen t geogr aphic partitions.In this case, a hash-based distributed overlay structure helps to main tain netw ork scalability and consistency .Faul t toler ance and load balancing are also handled by UbiFl ow.Besides, it provides visibility over under lying netw ork and optimizes the selection process of access poin ts within mul ti-netw orks so tha t QoS satisfie IoT da taf ow can be ensured.
How ev er, in the af oremen tioned works, the impact of SDN-sensed contextual inf orma tion in IoT resource discov ery has not been enlightened.Resource discov ery pla ys an importan t role in not onl y ensuring QoSsatisfie processing of IoT-da ta but also managing netw ork from being cong ested due uncertain load.Theref ore, the paper aims at SDN-based resource discov ery for IoT so tha t scalability in resource discov ery for IoT-da ta processing and forw arding can be ensured.

System Model
IoT-devices are geo-distributed and heterog eneous in terms of da ta sensing frequency and applica tionspecific tion.Due to energy constr ain t, IoT-devices cannot process any sensed da ta but using comm unica tion protocols like Constr ained Applica tion Protocol (CoAP), Sim ple Netw ork Manag emen t Protocol (SNMP), etc. can forw ard the sensed da ta tow ards Cloud or Fog for further processing.How ev er, here we assumed tha t, the IoT-devices and the computa tional entities can inter act through SDN.
Unlike traditional sta tic netw orking architecture (as shown in Fig. 1.a), in SDN (as shown in Fig. 1.b), da ta forw arding plane is decoupled from the controller plane.Here, a Centralized Controller componen t (CC) determines the routing pa th and da ta forw arding rules.The other netw orking entities like switches, gatew ays, access poin ts, base sta tions, etc. forw ards the da ta according the guidelines of the CC.
In order to iden tify the efficien t da ta routing pa th and computa tional entity, the CC senses the contextual inf orma tion and monitors the internal oper ations of netw ork, computa tional entities and IoT devices.In gener al, contextual inf orma tion provides enriched perception reg arding di fferen t system componen ts [12 ].Here, the contextual inf orma tion incl udes: -Curren t traffic load (netw ork throughput) on di fferen t routing pa ths.
-Curren t da ta processing load (size of queued da ta) on each computa tional entity.
-Data sensing frequency (da ta transmission rate) of IoT devices.
As the componen ts of the modelled system inter acts with each other through SDN, it is possible to tarck the context of each componen ts.Ref erence of sev eral context -sensing framew ork for SDN is available in the liter ature [13 ] [14 ].Any of the framew orks can be applied to track the af oremen tioned contextual inf orma tion of the computa tional entities, under lying ork and IoT devices.The sensed contextual inf orma tion helps CC to perceiv e the whole system efficien tl y and enhance the visibility over each of the componen ts.
Due to softw are define architecture, SDN can dynamicall y activ ate any idle computa tional entity and routes tow ards the entity whenev er the curren t system model becomes unable to meet the service demand.Moreov er, in an SDN-based system, an IoT-device is una ware about the computa tional entity where associa te IoT-da ta is going to be processed.In consequence, the system becomes able to provide virtualiza tion in processing IoT da ta and can be manag ed according to the dynamics of the environmen t.Hence, an SDN-based system supports scalability to a certain exten t.Conversel y, in the traditional sta tic netw ork architecture, IoT-da ta cannot be migr ated to other computa tional entity as it does not provide any virtualized settings.As a resul t it becomes very di fficul t to achiev e scalability in the traditional netw ork.
Necessary nota tion for modelling the system has been provided in Table .

SDN-based IoT-resource discovery
The proposed SDN-based IoT-resource discov er policy executes in the CC.Whenev er an IoT-device n sensed any da ta µ n from the external environmen t, it forw ards the da ta µ n through SDN to CC. Besides, the contextual inf orma tion of IoT-device n reg arding its da ta transmission rate λ n is also sen t to CC.Based on the receiv ed inf orma tion, CC runs the DiscoverResources proced ure as shown in Algorithm.1.
The DiscoverResources proced ure is consist of four basic steps.The steps can be describes as foll ows: 1.At firs , for each of the computa tional entity (line 4), it is checked whether the incl usion of µ n to its curren t da ta processing load exceeds the capacity of the corresponding computa tional entity (line 5).If it satisfies then the computa tional entity with minim um da ta processing load is selected as the targ et entity for processing µ n (line 6-8).This approach can be termed as the best -fi selection of computa tional entity.
2. La ter, from the available routing pa ths the suitable routing pa th tow ards the selected computa tional entity is iden tifie (line 10-13).In this case, the firs route is selected tha t cannot be cong ested due to per unit time da ta transmission from the IoT device n (line 11).This is considered as the firs -fi selection of the routing pa th.
3. In this step the sensed da ta µ n of IoT-device n is forw arded tow ards the selected computa tion entity through a cong estion-less routing pa th (line 14-16).
4. If no feasible computa tional entity or routing pa th is found, CC can dynamicall y initia te any idle if φ e + µ n < α e then 6: if φ e < η then Forw ard µ n to a e through a p computa tional entity and iden tify route tow ards the entity so tha t the sensed da ta µ n can be forw arded for processing.
The DiscoverResources proced ure combines best -fi and firs -fi selection approach (step 1-2) within it.Gener all y, the complexity of this al gorithm will increase linear ly as the number of computa tional entity increases.How ev er, due to step 1 and 2, it becomes easier to iden tify appropria te computa tional resources and associa ted routing pa th (step 3).Moreov er, due to basic features of SDN, it is also possible to accommoda te increasing service demand to idle computa tion entities (step 4).As a resul t, scalability issues in resource discov ery for IoT become attainable.
Since the DiscoverResources proced ure facilita tes resource discov ery and scalability , it pla y crucial role in minimizing resource discov ery time and in enhancing QoS satisf action for increasing number of IoT-service requests.Besides, not onl y in IoT, the proposed SDN-based approach can be extended to any sort of oper ations [15 ] where real-time inter actions are involved.

Performance Evaluation
In order to claim the feasibility of the proposed SDNbased IoT-resource discov er policy , at firs , the system has been sim ula ted and la ter the experimen tal resul ts are anal ysed.

Simulation Environment
The system model has been sim ula ted using iFogSim [16 ] sim ula tion toolkit.iFogSim sim ula tion toolkit has been dev el oped upon the CloudSim framew ork which has been used extensiv el y to sim ula te Cloud, Mobile Cloud, Vehicular Cloud environmen t.
In the sim ula tion, Fog nodes are considered as the computa tional entities and CC is a specialized Fog node to cond uct basic oper ations on SDN.In the modelled sim ula tion environmen t, IoT-devices can be placed at any location and the devices can ask for processing their sensed da ta by foll owing poisson distribution.
As the compa tible real-world workl oad is not curren tl y available, in the sim ula tion, syn thetic workl oad has been used.The workl oad and modelled system can be easil y re-constructible.Sim ula tion par ameters and units are represen ted in Table .2.

Simulation Results
The required time for iden tifying suitable computational resources is considered as one of the perf ormance metrics.In order to model resource discov ery time Eq. 1 has been applied.Here, the summa tion of da ta propag ation time (δ t ) from source IoT device to targ et Fog node and waiting time (ν t ) in Fog node has been iden tifie as total resource discov ery time (RD t ) for a da ta processing request.
Fig. 2 depicts tha t, resource discov ery time for IoT in sta tic netw ork is higher compared to SDN-based policy .Although in SDN-based approach, a certain amoun t of time is required by CC to iden tify appropria te targ et Fog node and the associa te routing pa th, the policy helps to red uce da ta processing waiting time and da ta propag ation time to a grea t exten t.In fact , the SDN-based sol ution selects tha t Fog node and tha t routing pa th as processing and comm unica tion medium in which processing load and netw ork cong estion is compar ativ el y less.Tha t's wh y in SDN-based sol ution resource discov ery time gets minimized.Conversel y, in sta tic netw ork based approach neither da ta processing overhead of Fog nodes nor netw ork cong estion is taken In addition to resource discov ery time, the percen tag e of QoS-sa tisfie da ta processing requests is considered as another perf ormance metric.Here, the deadline satisfie service deliv ery is taken into accoun t as a QoS par ameter .A da ta processing request satisfie QoS when the foll owing condition is satisfied here ∆ t is the service deliv ery deadline, τ t is the service response time.

∆ t > τ t
(2) Fig. 3 represen ts tha t, the percen tag e of QoS satisfie service requests in sta tic netw ork decreases significa tl y as the number of service requests increases.In SDN-based IoT, as scalable resource discov ery for increasing number of service requests is ensured, the percen tag e of QoS satisfie service requests alw ays remains in high.How ev er, a very less amoun t of downfall in the percen tag e of QoS satisfie service request is also experienced in SDN-based IoT as the number of service request rises.It happens due to run time activ ation of idle Fog nodes to meet the service demand.The required time for activ ating an idle Fog node has adv erse effect of the QoS satisfie service deliv ery of some requests.

Conclusion
The domain of IoT is expanding at a grea t pace.It is also experiencing di fferen t type of challeng es in its way of pr actical applicability .We have targ eted one of such challeng es of IoT in respect of scalable resource discov ery.Here, the proposed SDN-based resource discov er policy for IoT uses contextual inf orma tion of computa tional entities, netw orks and IoT devices to iden tify suitable resources and routing pa th to process and forw ard IoT-da ta.The policy is independen t of increasing number of da ta processing (service) requests tha t comes from geo-distributed IoT-devices.In consequence, the policy facilita tes scalable resource discov ery in IoT.Moreov er, sev eral sim ula tion studies also claim the feasibility of the proposed policy in respect of resource discov ery time and QoS satisf action rate of service requests.The SDN-based sol ution is substan tiall y efficien t compared to the sta tic netw ork based resource discov ery for IoT.
In future we aim at extending SDN-based sol utions to other aspects of IoT such as SDN-based IoT netw ork manag emen t, SDN-assisted conten t distribution in IoT, applica tion depl oymen t in SDN-enabled IoT.

4
EAI Endorsed Transactions on Scalable Information Systems 07 2017 -09 2017 | Volume 4 | Issue 14 | e4 Software Defined Network-based Scalability in Resource Discovery for Internet of Things

Figure 3 .
Figure 3. Percentage of QoS satisfied service requests vs number of service requests

1 Table 1 .
Notations n Sensed da ta by any IoT device n.
Iden tify route a p tow ards a e ; a p ∈ P