1 Fog-Assisted wIoT: A Smart Fog Gateway for End-to-End Analytics in Wearable Internet of Things Nicholas Constant*, Debanjan Borthakur*, Mohammadreza Abtahi, Harishchandra Dubey, Kunal Mankodiya Department of Electrical, Computer and Biomedical Engineering, University of Rhode Island, RI, USA Email: [email protected], [email protected] *Authors contributed equally 7 1 0 2 Abstract—Today, wearable internet-of-things (wIoT) devices before the cloud must be adapted to better utilize the cloud n continuously flood the cloud data centers at an enormous rate. services available. That is why in this paper we will discuss a This increases a demand to deploy an edge infrastructure for the development of a smart fog to help mitigate the amount J computing, intelligence, and storage close to the users. The 5 emerging paradigm of fog computing could play an important of data sent to the cloud by playing the role of orchestrator 2 roletomakewIoTmoreefficientandaffordable.Fogcomputing in the process of data acquisition, conditioning, analysis, and is known as the cloud on the ground. This paper presents an short-term storage. Our discussion will start with an overview ] end-to-end architecture that performs data conditioning and ofthefogenvironmentbydefiningtherolesofanIoTdevice, C intelligentfilteringforgeneratingsmartanalyticsfromwearable fog node, and cloud. This will transition into a description of D data. In wIoT, wearable sensor devices serve on one end while the communication protocols used between the groups, along thecloudbackendoffersservicesontheotherend.Wedeveloped s. a prototype of smart fog gateway (a middle layer) using Intel with the fog setup used during the benchmarking. Finally, we c Edison and Raspberry Pi. We discussed the role of the smart comparetheresultsofourbenchmarkingexperimentstothose [ fog gateway in orchestrating the process of data conditioning, of realistic payloads to explore limitations in scale. intelligent filtering, smart analytics, and selective transfer to the 1 cloudforlong-termstorageandtemporalvariabilitymonitoring. II. RELATEDWORKS v A. Wearable Internet-of-Things - wIoT We benchmarked the performance of developed prototypes on 0 real-worlddatafromsmarte-textilegloves.Resultsdemonstrated IoT or internet of things is the internet working of physical 8 theusabilityandpotentialofproposedarchitectureforconverting devices or other items embedded with electronics, software, 6 the real-world data into useful analytics while making use of sensors and network connectivity. IoT enables the connected 8 knowledge-based models. In this way, the smart fog gateway 0 objects to collect and exchange data [2]. We define a new enhances the end-to-end interaction between wearables (sensor . segment of IoT that is named as wearable IoT (wIoT) [3]. 1 devices) and the cloud. Wearable devices are rapidly emerging and this gives rise to 0 7 the wIoT due to their capability of sensing, computing and 1 I. INTRODUCTION communication [3]. We also discussed the building blocks v: Wearable sensors continuously collect and stream data to of wIoT which includes wearable sensors, internet-connected i clouddevicesfordataprocessingandlong-termstorage.These gatewaysandcloudandbigdatasupportthatcanbeinstrumen- X wearable sensors are low-power, low-bandwidth devices, and tal in its future success in healthcare domain applications [3]. ar close proximity devices. In an effort to expand battery-life IoTistransforminglives.Wearabledevicesthatofferbiometric and response time of the sensors, the fog concept, often measurements such as heart rate, perspiration levels, oxygen referred to as edge computing, has grown more common- levels in the bloodstream are available even in the smartwatch place among system architectures. The fog concept coined itself. In [4] authors assert that technology advancements may by CISCO aims to minimize latency, conserve bandwidth, even allow alcohol levels or other similar measurements to be improve security, maintain reliability, and move data to the made via a wearable device. Complex sensors are increasing best place for processing [1]. This concept allows for more the efficacy of wIoT, like tracking body temperature might sensors,oredgedevices,tointerfacewiththecloudonamuch provide an indication of flu. Wearable devices connectivity larger scale than previously seen. Since the cloud is not setup with household appliances can make its work more versatile. for this volume and variety of data changes in the systems WIoT light sensors might be used for controlling the lighting in home etc. The privacy threat that might come into picture Thismaterialispresentedtoensuretimelydisseminationofscholarly while dealing with biometrics associated with wIoT can be and technical work. Copyright and all rights therein are retained by an interesting issue to tackle in near future. Nobody wants to theauthorsorbytherespectivecopyrightholders.Theoriginalcitation of this paper is: N. Constant , D. Borthakur, M. Abtahi, H. Dubey, K. sharetheirbloodpressurelevelstoanunknownperson,unless Mankodiya,"Fog-AssistedwIoT:ASmartFogGatewayforEnd-to-End he or she is their doctor. For smooth working of high-end IoT Analytics in Wearable Internet of Things", The 23rd IEEE Symposium applications, significant processing and interface capabilities on High Performance Computer Architecture HPCA 2017, (Feb. 4, 2017 âA˘S¸ Feb.8,2017),Austin,Texas,USA. are always needed. 2 Fig.1. Theconceptualoverviewofthefog-assistedwIoTarchitectureforend(sensors)-to-end(cloud)analytics.Thedatageneratedinwearablesisconditioned withinfogdevice.Inaddition,temporarystorageandintelligentfiltering,analysisandpredictionisperformedondata.Thefinalanalysisweresenttocloud backendforlong-termstorageandglobalanalysis. B. Management of Big Data in Wearable Internet of Things fog computing can improve the efficiency and performance of application [2]. In [9] authors emphasize on the increasing Widespread use of wearable and internet of things have need for implementing machine learning algorithms includ- leadtoseveralinterestingapplicationsandalsounprecedented ing deep learning on resource constrained mobile embedded challenges. For enhancing the data management and analytics devices with limited memory and computing power. Authors in wIoT, edge computing have emerged. Fog computing is used a 2 Bit network for model compression, this achieves emerging as a tool for such situations. The data generated by a good trade-off between model size and performance. The wearables are of temporal and spatial nature. Employing edge scopesandopportunitiesofFogcomputingisenumerable.Fog devicesforanalysisandvisualizationofdataleadstoefficient computing supports a growing variety of applications such as solution leading to improvement in overall power efficiency. those in the Internet of Things(IoT), Fifth generation (5G) The big data being generated from various applications could wireless systems, and embedded Artificial Intelligence (AI) be explained by four Vs namely volume, velocity, variety and [10]. veracity [5]. The harmony between wIoT and big data could lead to generating valuable analytics from big data. Fog com- III. THEARCHITECTUREOFSMARTFOGGATEWAY putingholdsagreatpromisetoreducetheburdenofwearable ThefogenvironmentconsistsofIoTdevices,fognodes,and big data at the edge of the network. Authors in [6] propose a the cloud. The fog is aimed at alleviating strain on the cloud novel BigEAR big data framework, where it identifies mood by sharing summarized activities of the IoT devices, as well ofthepersonfromvariousactivitiessuchaslaughing,singing, as enacting the learned rules created on the cloud. In order crying,arguingandsighing.OurproposedFoggatewaycanbe to develop the context in which the fog node is applicable, made to incorporate such versatile clinical speech processing the key roles for the groups identified in Figure 1 will be framework. Such framework is well accounted in literature explained. suchasin[7],whereauthorspresentaFogcomputinginterface [7] for processing clinical speech data. A. Smart Fog Computing: Semantics, Cognition and Percep- tion C. Fog Computing: Architecture, Feasibility and Opportuni- We used knowledge-based models for computation and ties analytics from big data collected by wearables connected to Fog computing as defined in [8] is a model to complement internet. Smart IoT-Based Intelligent Perception is emerging the cloud for decentralizing the concentration of computing as a new service-oriented model. Authors in [11] propose a resources (for example, servers, storage, applications and five-layered structure (i.e., resource layer, perception layer, services) in data centers towards users for improving the network layer, service layer, and application layer) resource quality of service and their experience. Fog computing [8] intelligent perception and access system based on IoT. It is the process of decentralization of computations which is not only provides promising attention to healthcare but also away from the cloud and towards the edge of the network in powerful industrial systems utilizing the ever growing closer to the user. In this way, fog computing increases the wireless, sensor devices and embedded systems. Social IoT quality of service and reduces the latency and frequency of came which encompasses social networks with IoT. Intelli- communication between a user and an edge node. Moreover, gent perception and making use of different sensing devices 3 and adapters Fog actually mimics human perception. Sensing on the use of Bluetooth among IoT devices and fog nodes. devices might include barcode, RFID readers, sensors, GPS, The communication within IoT devices needs to be consistent etc. The adapter includes software interface adapter, sensor overalongrangewhilemaintaininglowpower.Thebandwidth adapters,modeladapters,etc.SmartFogdevicesmightbewell required for the various devices depends on the application, configured to be adaptive to real-time data streams. They can although our wearables are satisfied with the far less than be modeled with algorithms that will make their processing 25Mbps as provided by Bluetooth 4. energy efficient, encrypted and with low latency, and hence the name ’smart’ aptly befits the Fog gateway proposed in IV. RESULTS&DISCUSSIONS this paper. A. System Description The Intel Edison platform used in this analysis has a dual- core, dual-threaded Intel Atom CPU at 500MHz with a 32-bit B. IoT Device Intel Quark microcontroller working at 100MHz, Bluetooth The IoT devices interacting with the fog node will mini- 4.0anddual-bandIEEE802.11a/b/g/nviaanon-boardchip mally consist of sensors or actuators capable of transmitting antenna. Yocto is the Linux environment for this plateform, a collected data via Bluetooth. These devices may be limited prebuilt distribution of Debian/Jessie for 32-bit systems was in power, processing and storage. The topology among the deployed as Yocto is not an embedded distribution of Linux devices can vary, but in this paper we focus on a mesh itself. It only provides an environment to develop a custom topology. The main role of the IoT device is to sample real Linux distribution. This was done so that same environment worlddata,whenneededtransmitthatdatatothefognodefor onboththeIntelEdisonandtheRaspberryPicanbeachieved. decisionsandstorage.Thesedevicesmaysampleathighrates The Raspberry Pi Model B platform used in this analysis for applications such as biological sensing, down to low rates has a system consisting of a 900MHz 32-bit quad-core ARM for applications such as monitoring contents within a fridge. Cortex-A7 CPU, with 1GB RAM. A WIFI dongle based on theRealtekRTL8188CUS chipsetwasinstalledasRaspberry C. Fog Node Pi does not have WIFI connectivity. Raspbian is the custom Linux distribution for this platform. Raspbian was replaced The role for a fog node is to orchestrate communication with the Debian/Jessie distribution used on the Intel Edison withtheIoTdevicesanddeliverreal-timeresponsesbasedoff for computational flexibility. of rules determined by the supervised learning. The fog node will condition, analyze, and package this data for the cloud to learn and store. The fog node can use a variety of software B. Smart Glove Data for the cleaning and learning, but this paper will focus on Smart Glove [12] is a wearable, completely wireless device the use of Octave. The nodes and IoT devices are configured transmitting the data recorded by the microcontroller Arduino in a self-healing mesh topology using a routing technique. 101 to a smart phone or computer via Bluetooth. The sensors Communicationtothecloudisaccomplishedviaweb-sockets. that are integrated in the Smart Glove are Spectra Symbol This paper focuses on the use of Socket.io for Node.js. flexsensorsthathaveathicknessof6.35mm,with84.86%of the part length designated as active length. The flex sensors D. Cloud are analog resistors which act as variable analog voltage The cloud acts as the overseer for the overall system by dividers. The voltage across the flex sensors will change if employing a combination of deep learning, long-term storage, they are bent, therefore, we can convert it to the resistance of and more. However, this paper focuses on the use of Ten- the flex sensor that changes, and the aim is to measure the sorFlow for deep learning. The cloud can disperse the latest angular displacements where the flex sensor is located. In our updates down to the IoT device via the fog node. The cloud prototype design of the Smart Glove, we have used two flex is not required to deliver real-time responses down to the fog. sensors located on pointing finger and thumb (see Figure 2 in Instead, the intention is to migrate that responsibility to the order to quantify how much these fingers are bent during the fog node. experimentoffingertappingwhichisacommonpracticeinthe treatmentprocedureofParkinson’sdisease.Inthisexperiment, theSmartGlovedatacanrevealhowmuchtheparticipanthas E. Communication Protocols Used bent their fingers to do the finger tapping, and also show how 1) Bluetooth 4: Between IoT devices and Fog Nodes fast they can do it. Figure 3 shows the frequency of finger 2) TrickleAlgorithm:Workaroundtocreatemeshtopology tapping in an experiment that the participant was asked to using Bluetooth 4 start finger tapping slowly, and gradually doing it faster at 3) Socket.io: Between Fog Nodes and Cloud different rounds, and in the last round, were asked to start A key component to in a fog environment is the ability slowly and getting it faster. Therefore, the Smart Glove can for continuous communication in directions down to the IoT be used as a wIoT device in the movement therapy of any devices and up to the cloud. There are currently a variety of kind of disorders that needs monitoring of the hand motion, wireless protocols based off of the IEEE802.15.4 such as such as Parkinson’s disease. The patients can wear the Smart Thread,ZigBee,Bluetooth,andmore.Eachoftheseprotocols Glove at home, doing the assigned exercises and the data can allow for mesh topologies. Although we will focus primarily besenttothephysician,withouthavingtomakeatriptovisit the doctor. 4 servers could be summarized from Figure 3 under total run- time. Little’s Law can be used to determine the average wait time for each device, that is LeadTime=WIP(units)/ACR(unitspertimeperiod), where WIP is Work-in-Process and ACR is Average Comple- tion Rate for any process [13]. While running one set at a time, the average wait time for the Intel Edison and Raspberry Pi was around 64.65 seconds and 12.39 seconds, respectively. Both of the devices used as fog servers were used to collect data from the smart glove, smartwatch, wristband and/or other wearable devices, then next step was data processing and finally, the processed data was sent to a server in the cloud. Figure 4 shows the average Fig. 2. Illustration of Smart Glove showing the microcontroller Intel Curie andthelocationofflexsensors. amountoftimespentoneachstep.TheRaspberryPiprovided a service rate that is one-third of that of Intel Edison. Also, it was found Raspberry Pi consumes 198mW/s and the Edison consumes 529mW/s when active. Fig.3. FrequencyoffingertappingresultedfromtheSmartGlove.Rounds 1 and 2 are tapping slowly, round 3 getting faster and round 4 is as fast as possible,round5startswithtappingslowlyandgraduallygettingfaster. C. Benchmarking and Program Setup The Fog devices were logged into with the SSH protocol. Some Benchmarking scripts were then run. The scripts starts Octave and load it with the data. The script searches for the process ID (PID). Once determined the PID, it would extract all the information top provided about the systems performance and also the load on the system by this instance of Octave. The information was logged into a .csv file and Fig. 4. Performance comparison between Intel Edison and Raspberry Pi in saved for analysis. Profiler function in the background starts proposed architecture. We used real-world smart-glove data for this bench- marking experiment.The average load put onto the board when processing once the instance of Octave was ready to run the algorithm. isshownhere.AlsoMemoryLoadandCPUloadareshownaspercentages At the end, the Profilers set of data was stored into a .mat basedondata-set. file for later analysis. D. Analysis E. Limitations The total process breakdown shown in Figure 4 includes Cloud and Fog always play a mutually beneficial inter- the load added on by starting an instance of Octave. The run dependent relationship. The limitations might encompass sev- time for the Intel Edison was much greater than that of the eral areas, one of them is the Cloud-Fog Interface. Fog can Raspberry Pi and an increase in data sets (size N) produces not provide the massive storage and complex heavy computa- a processing time of the order of Nlog(N). It is noticed that tions and the problem of wide area connectivity is another Raspberry Pi could complete the process almost two times issue. Another limitation may be the security. Authors in faster and it could scale to 125+ datasets where as the Edison [10] mention that, though fog might enhance security but it might crash. The network created by us places the Fog server might pose some new security challenges while dealing with inbetweentheSmartGlove(orotherwearables)andthecloud. privacy sensitive data. Other privacy issues that might include It was assumed that the mean arrival rate of data would be data privacy, location privacy, usage privacy i.e. the usage once per minute assuming the Fog device would be placed pattern with which a fog client utilize the fog services. In the in locations where only a small number of devices in the paper [14] authors give an example of it, in smart grid, the area exist, such as a hospital. The service rate for the Fog reading of the smart meter will disclose lots of information 5 of a household, such as at what time there is no person at [16] H. Dubey, J. Yang, N. Constant, A. M. Amiri, Q. Yang, and home,andatwhattimetheTVisturnedon,whichabsolutely K. Makodiya, “Fog data: Enhancing telehealth big data through fog computing,”inProc.ASEBigData&SocialInformatics2015. ACM. breaches user’s privacy. Inadequate storage and inability to do [17] H. Dubey, A. Monteiro, L. Mahler, U. Akbar, Y. Sun, Q. Yang, and highperformancecomputingisanotherareaofconcernforfog K. Mankodiya, “Fogcare: A fog-assisted internet of things for smart devices.Theresearchpresentedinthispaperutilizedaclunky telemedicine,” in In the International Journal of Future Generation ComputerSystems,Elsevier.,2017. mesh network for the IoT devices. In future works, Bluetooth [18] H. Dubey, J. C. Goldberg, M. Abtahi, L. Mahler, and K. Mankodiya, 5 will be used which will provide a proper mesh network for “Echowear:smartwatchtechnologyforvoiceandspeechtreatmentsof these devices. patientswithparkinson’sdisease,”inProceedingsoftheconferenceon WirelessHealth. ACM,2015,p.15. V. CONCLUSION Massiveamountsofdataisbeinggeneratedwithwidespread use of wearables, sensor devices in a variety of applications such as healthcare, smart-city, smart-grid, assisted homes etc.. We proposed, developed and validated a smart fog gateway thatcouldperformend-to-endanalyticsforinternet-connected wearables. Comparison between Intel Edison and Raspberry Pi provide evidence on choosing embedded boards that could match the demands of the application. The benchmarking experiments were done on real-world data. REFERENCES [1] CISCO, “White paper on fog,” https://www.cisco.com/c/dam/en_us/ solutions/trends/iot/docs/computing-overview.pdf. [2] “Internet of things global standards initiative.” [Online]. Available: http://www.itu.int/en/ITU-T/gsi/iot/Pages/default.aspx [3] S.Hiremath,G.Yang,andK.Mankodiya,“Wearableinternetofthings: Concept, architectural components and promises for person-centered healthcare,” in IEEE Wireless Mobile Communication and Healthcare (Mobihealth)Conference,2014. [4] M.Electronics,“Wearabledevicesandtheinternetofthings.” [5] Y.Sun,H.Song,A.J.Jara,andR.Bie,“Internetofthingsandbigdata analytics for smart and connected communities,” IEEE Access, vol. 4, pp.766–773,2016. [6] H. Dubey, M. R. Mehl, and K. Mankodiya, “Bigear: Inferring the ambient and emotional correlates from smartphone-based acoustic big data,” in Connected Health: Applications, Systems and Engineering Technologies (CHASE), 2016 IEEE First International Conference on. IEEE,2016,pp.78–83. [7] A.Monteiro,H.Dubey,L.Mahler,Q.Yang,andK.Mankodiya,“Fit:A fogcomputingdeviceforspeechtele-treatments,”inSmartComputing (SMARTCOMP),2016IEEEInternationalConferenceon. IEEE,2016, pp.1–3. [8] N. Wang and Buyya, “Feasibility of fog computing,” https://www.researchgate.net/publication/312550677_Feasibility_ of_Fog_Computing. [9] W. Meng, Z. Gu, M. Zhang, and Z. Wu, “Two-bit networks for deep learning on resource-constrained embedded devices,” arXiv preprint arXiv:1701.00485,2017. [10] M. Chiang and T. Zhang, “Fog and iot: An overview of research opportunities,”IEEEInternetofThingsJournal,vol.3,no.6,pp.854– 864,2016. [11] F.Tao,Y.Zuo,L.DaXu,andL.Zhang,“Iot-basedintelligentperception and access of manufacturing resource toward cloud manufacturing,” IEEETransactionsonIndustrialInformatics,vol.10,no.2,pp.1547– 1557,2014. [12] V.C.M.B.N.C.BaharFarahani,FarshadFirouziyandK.Mankodiya, “Towards fog-driven iot ehealth: Promises and challenges of iot in medicine and healthcare,” International Journal of Future Generation ComputerSystems,2017. [13] “Little’s law.” [Online]. Available: https://www.isixsigma.com/ dictionary/littles-law/ [14] S.Yi,Z.Qin,andQ.Li,“Securityandprivacyissuesoffogcomputing: Asurvey,”inInternationalConferenceonWirelessAlgorithms,Systems, andApplications. Springer,2015,pp.685–695. [15] A. Sheth, “Internet of things to smart iot through semantic, cognitive, andperceptualcomputing,”IEEEIntelligentSystems,vol.31,no.2,pp. 108–112,2016.