A Novel Big Data Analytics Framework For Smart Cities - ترجمه یار

2.1.1 Big data an nalytics platfforms a tools req quired for kknowledge discovery d Big ddata analyticcs refers to the entire processes and includding data exxtraction, transformationn, loading and a analysis; specific toools, techniq ques, and methoods; and how w to successffully providee results to deecision makeers. Althoughh developing g big data analyytics platform ms encounterr non-trivial cchallenges due d to the com mplex naturee of big data, big data analyytics holds an a unpreced dented opporrtunity to shift the trad ditional metthods of infformation extracction into neew dimensions. This oppportunity ind duced researcchers and tecchnology pro oviders to devellop sophisticcated platform ms, framewoorks and algo orithms to compete with tthe challengiing of big data [[6, 11, 12]. To deeal with the challenging nature of bigg data, platfo orm scalabiliity representss the logical solution. Theree are two com mmonly scalling approacches: verticall and horizon ntal scaling w which are kn nown also as sccale up and scale out reespectively [6]. Verticall scaling meeans empow wering the prrocessing platfoorm with addditional computing poower (memo ory, CPUs etc.) to acccommodate with the increm mental volum mes of data. This appro ach involvess execution of a single ooperating sy ystem. On the otther side, horrizontal scaliing is a dividde-and-conqu uer approach h. The worklo load is distrib buted and proceessed in parallel across multiple inddependent computing machines. m Moore machinees can be addedd as much as a needed to o improve thhe overall system s perfo ormance. Thi his approach involves execuution of multtiple instancees of differennt autonomo ous operating g systems runnning on ind dependent machhines, which as such a furrther compleexity (Figure 2). Figure F (2) - V Vertical vs Horizontal H Scaling Each approach haas its advantages and dissadvantages. Although veertical scalinng shows resilience in platfoorm upgradiing, yet, it is i restricted by the plattform upper limits (e.g. maximum memory, numbber of CPUss etc.). On the other siide, in horizzontal scaling g more com mputing nodees can be addedd as much ass needed. Ho owever, horiizontal scalin ng involves maintaining m multiple insstances of differrent operatinng systems th hat is a com mplicated chaallenge. High h Performancce Computin ng (HPC) clusteers and Apacche Hadoop [13] [ are two examples off vertical and d horizontal sscaling platfo orms. Howeever, the cosst of verticall scalability and upper ceiling c limitaations are maajor and con nsiderable drawbbacks of verrtical scaling g. These twoo drawbacks come in fav vor of horizoontal scaling g when it comees to smart city projectss, considerinng the dynam mics of mullti-domain nnature of sm mart cities projects and prosspects for futture expansioon, it is morre rational to o rely on horrizontal scaliing rather than vertical scalling. This note n interpretts why mosst of the researches andd designs of big data analyytics framewoorks and plaatforms are bbuilt using ho orizontally sccalable platfo forms such as Apache Hadooop platform. 2.1.2 Big data vaalue chain op perators ogues approaach used forr Knowledgee Discovery in Databasees (KDD) In [6], authors addopted analo nding bottlennecks in casse of big data analytics. In the KDD D model, modeel to study thhe correspon analyytics is divideed to three operators: inpput, analysis and output (Figure ( 3). Inn the case off big data, data input operaator ingest stream s volum mes of noisy incomplette raw data from heterogeneous sourcces. The proocesses of in nput operatorr have a cru ucial impact in mitigatinng the effecct of data

volum minous on the t overall analysis proocesses. Piccking up rellevant data, data clean nsing and comppression will influence th he efficiencyy of the dataa analysis execution perfo formance. Deespite the importance of thee input operaator processees, the autho ors in [6] notted an imporrtant observaation that the nuumber of ressearch articlees and techniical reports th hat focus on data analysiss operator is typically more than the num mber focusin ng on other ooperators. Figu ure (3) - Knoowledge Disccovery in Daatabases a the survey in [66] revealed that t parallel computing aand cloud co omputing As foor the data analysis, technnologies havee a strong im mpact on the rresearch on this t area, wh here most of the big data analytics frameeworks and platforms were w developped based on o these tech hnologies. H However, relliance on paralllel distributeed processin ng platformss compelled d researcherss to modifyy traditional machine learniing and datta mining algorithms a too adapt witth the new environmennt and prog gramming paraddigm (e.g. Map and Redu uce) [13]. 2.2 Smart cities Althoough the term m smart city seems to be simple and intuitively i un nderstandablee, there is no o globally recoggnized definiition of the term; this iis due to diffferent persp pectives abouut the how the label “smaart” is loadeed in the co ontext of sm mart cities [1]. [ There are a several surveys in literature addreessing the deefinition of “smart city” ffrom differen nt perspectiv ves [1, 2, 3, 55]. For exam mple, IBM definnes a smart city c as: “thee city that m makes optim mal use of alll the intercoonnected infformation availaable today to t better und derstand andd control its operations and optimizze the use of o limited resouurces” [5]. Also, A YIN Ch huanTao, et al [3] defineed a smart city as “a sysstem of tech hnological infrasstructure thatt relies on ad dvanced dataa processing with w the goal of making ccity governaance more efficiient, citizenss’ happier, business b moore prospero ous and the environmennt more susttainable”. Authoors in [2] annalyzed and categorized several defiinitions of th he term “sm mart city” in literature settinng three fuundamental factors thatt make a city smart. These facttors are technology (infraastructures of o hardware and softw ware), peoplee (creativity, diversity, and educattion) and instituution (goverrnance and policy), p giveen the conneection betweeen these facctors. Howev ver, from these definitions we w can draw w out three esssential charaacteristics thaat characterizze smart citiees as: Thhe vital rolee of ICT ass key enablling technolo ogy in deveeloping smar art cities, wh here ICT represents the essential baackbone for connecting city core sy ystems togethher infusing data and innformation beetween diffeerent city sysstems. Digitaal artifacts an nd ICT appliications generate raw daata that can be b blended and a analyzedd to extract useful u inform mation and innsights abou ut the city ussing artificiall intelligencee and data annalysis appliccations. Thhe integral view v of a sm mart city, whhere the interrrelationship between thee core system ms of the citty should bee considered d. Scholars aand relevantt organizatio ons define diifferent dom mains that coomprise a sm mart city, forr example, aauthors in [4 4] defined six x smart dom mains for a city c to be sm mart. In this regard, r a sm mart city is vieewed as a whole w body off systems (i.ee. system off systems) whhere no systeem in the city y domains fuunctions in issolation [14, 15, 16]. Suustainability,, which meaans in its brroad meanin ng the ability y to continuue and grow w without siggnificant detterioration. In n smart city context, the concept of su ustainabilityy applies into o different asspects of life in a smart ciity such as ecconomics an nd environmeent.

Thesee characterisstics are not exclusive to smart cities, whereas there are manyy other relatiive labels in literature suchh as “digital city”, “inteelligent city””, “virtual ciity” and “ubbiquitous city y” which addreess different aspects thatt characterizze modern cities and shaare some off these charaacteristics with smart cities with w differen nt degrees off deepness. 2.2.1 Modeling of o smart citiees - an ICT perspective From m an ICT perrspective, ind dustrialist annd scholars adopted a the layered apprroach to mod del smart citiess. For exampple, IBM adopted a three--layer modell including in nstrumented layer, intercconnected layer and intelliggent layer [5]. Similarlyy, YIN Chu uanTao, et al. a proposedd a four-layer model includding: data accquisition an nd transmissioon layer; datta vitalization n layer; com mmon data an nd service layer; and the appplications lay yer [3]. Reggardless of th he number of layers, thesse models prroject the journney of the biig data from its birth in its raw form m until valuaable informaation and inssights are extraccted for the benefit b of deecision makerrs and citizen ns (Figure 4)). Although tthere is no on ne-to-one correspondence between b big data value chain operaators (Figuree 3) and the layers of smart city n objective off projecting the t processes of extractinng insights and a useful modeels, they sharre a common inform mation out of o raw big datta to supportt smart servicces and decission-making processes [1 17]. Figure (4) - Big Data in i Smart Cities 2.3 B Big data anallytics in sma art cities The rrole of big daata in developing smart ccities is undeniable [7]. There T are manny applicatio ons of big data analytics inn different domains d of smart citiess, such as planning p [8]], traffic con ntrol and transpportation [188, 19], crimee analysis [200], energy [2 21] and envirronment [22]]. To design software platfoorms and arrchitectures fitting smarrt city purpo oses, specifiic non-functtional and functional f requirements linkked to smarrt city’s natuure of data sources and d applicationns should be b clearly identiified first. Foor example, the heterogeeneity of data sources (e.g. sensed IooT data, Sociial Media Netw work and Ellectronic Meedical Recoords) entails considerable design ffactors such h as data integrration, system m scalability y, privacy annd security [9 9, 23]. Also, the dynamicc nature of smart s city life nnecessitates considerablee attention ffor stream data d analyticss to enable online and real-time servicces [8, 9]. Additionally, A “historical ddata” or “battch data” anaalytics is an essential req quirement for pllanning (shorrt and long teerm) and deccision-makin ng purposes in i smart citiees. In [17], authors compiled a comprehenssive list of eleven e essen ntial requirem ments that should s be consiidered in dessigning smarrt cities softw ware architecctures. Typiccally, these reequirements could be classiified into twoo categories. Firstly, the functional requirements: object interroperability, real time monitoring, histoorical data, mobility; m servvice compossition and integrated urbban managem ment. The seconnd category is the non--functional rrequirementss: sustainability, availabbility, privaccy, social aspeccts and flexibbility/extensiibility (scalabbility). Simillarly, authorrs in [24] introduced a ccomprehensiive study baased on exam mining 23 smart city platfoorms to specify the fun ndamental fuunctional and non-functional requireements to develop d a

software platform to enable construction of scalable integrated smart city applications. The study concluded with eight functional requirements: data management, application run-time, Wireless Sensor Network WSN management, data processing, external data access, service management, software engineering tools and definition of city model. This is in addition to eight non-functional requirements: interoperability, scalability, security, privacy, context awareness, adaption, extensibility and configurability. Authors in [23] specified six characteristics a big data analytics platform should maintain to accommodate with the V challenges of big data, specifically: scalability, I/O performance, fault tolerance, real-time processing, data size and support for iterative tasks. In summary, we can recap the functional requirements into interoperability, real time analysis, historical data analysis, mobility, iterative processing, data integration and model aggregation. While the non-functional requirements are scalability, security, privacy, context awareness, adaption, extensibility, sustainability, availability, and configurability. 3. The Scope of Literature Review In order to define the scope of this literature review, authors followed a widely known taxonomy scheme proposed by [25] and adapted by [26] that includes six characteristics for literature review: (1) focus, (2) goal, (3) organization, (4) perspective, (5) audience and (6) coverage. 3.1 Focus Focus is the central area of interest to the review process. According to [25], it could be research outcomes, research methods, theories, practices or applications. The candidate articles for review are analyzed in two dimensions. The first is the analysis with respect to big data value chain operators presented in subsection 2.1.2 (appendix B). The second one is the analysis with respect to the functional requirements mentioned in section 2.3 (appendix C). The reason for choosing these two dimensions for analysis is that they complement each other. This approach gives a broader picture about the traits of big data analytics frameworks in smart cities, which in return gives a pointer to answer the first research question. In this regard, it is worth mentioning that nonfunctional will be not be addressed in this article as it is considered general requirements applicable for any analytics system which also required in smart cities with different levels of complexity. 3.2 Goal Goal refers to the objectives to be fulfilled by the review that could be integration, criticism, central issue. As the objective of this research is to study the traits of the available big data analytics frameworks applied in smart cities and identify the essential functionalities that should characterize big data analytics frameworks to serve smart cities purposes, the goal of this research is to integrate and criticize the finding of the past literature. 3.3 Organization Organization refers how the literature review is organized. The literature could be organized as chronological order, conceptual order (sharing of the same ideas) or methodological order (sharing of the same methods of work). In this article, literature is organized and discussed in a conceptual order. 3.4 Perspective Perspective refers to the reviewer’s point of view in discussing the literature. Perspective could be: a neutral position (impartial role as an honest “judge”) or an espousal position (advocate to certain

idea(s) or methodology). In this research, the author adopted a neutral position search perspective since there is no need to foster a specific position. 3.5 Audience Audience refers to beneficiaries whom the review addresses (specialized researchers, general researchers, practitioners, policy makers). Since the second research question is identifying essential functionalities that should big data analytics frameworks have to serve smart cities purposes, the audience of this literature review are specialized scholars, practitioners and smart cities planners. 3.6 Coverage Coverage refers to how the reviewer searches the literature and how he makes decisions about the suitability and quality of documents. According to Cooper, there are four categories of coverage: exhaustive (including the entirety of literature on a topic or at least most of it), exhaustive with selective citation (considering all the relevant sources, but describing only a sample), representative (including only a sample that typifies larger groups of articles), and central (reviewing the literature pivotal to a topic). In this literature review, authors adopted exhaustive with selected citations coverage since it is not realistic to claim exhaustive coverage. Additionally, adoption of representation or central coverage does not serve the objectives of this review. Table (1) summarizes the choices made by the author, regarding the Cooper’s taxonomy about the review scope. Table (1) Taxonomy of literature review Characteristics Categories Focus Research outcomes Goal Integration Criticism Central Issue Organization Historical Conceptual Methodological Perspective Research methods Neutral representation Theories Applications Espousal of position Audience Specialized scholars General scholars Practitioners\ Policy-makers General public Coverage Exhaustive Exhaustive with selective citation Representative Central or pivotal 4. Literature analysis and synthesis According to the reference review scheme [26], the search process involves four steps: (1) identifying search databases; (2) search keywords; (3) forward and backward search; and (4) evaluation of articles. To collect quality scholar articles, six information systems online databases were searched using two search keywords “big data” and “smart city”, total of 247 articles were retrieved. After filtration and evaluation, process only 30 articles were shortlisted for analysis. Details of the search process and sources of shortlisted articles are listed in appendix A. The process of evaluating the articles in terms of addressing the two dimensions of analysis in section 3.1 involved reading and evaluating the article’s abstract and conclusion sections to decide which of the focus points (section 3.1) are addressed. If article abstract and conclusion does not

lead into clear decision, the article’s full body is reviewed. In the following two subsections, the results of analysis with respect to big data value chain operators and functional requirements are demonstrated. For a more detailed analysis of some proposed end-to-end architectural frameworks are reviewed in details, namely: BASIS [27], SWIFT [28] and RADICAL [16] are reviewed in details in subsections 4.3, 4.4 and 4.5 respectively. Points of strength and weakness of each architectural framework are listed after each review. 4.1 Analysis with respect to value chain operators Results of the evaluation process with respect to big data value chain operators are shown in the table presented in appendix B. From these, we could recognize that most of the analyzed articles focused on data gathering (21 out of 30). This ratio reflects the high interest of researchers in finding efficient solutions for data gathering from different sources, while relatively, less number of researches addressed the rest of data input functionalities (selection, preparation, transformation) although its significant impact on the efficiency of the analysis processes. In the following three subsections, we will review how each of these operators is addressed. 4.1.1 Data Input The central challenge in this operator is related to the ability to acquire timely raw data about city events from a large number of heterogeneous sources. Also, in through this operator data are prepared to the following operator for

The process of analyzing big data to extract useful information and insights is usually referred to as big data analytics or big data valu e chain [6], which is considered as one of the key enabling technologies of smart cities [7, 8, 9]. However, big data complexities comprise non-trivial challenges for the processes of big data analytics [3].