计算机专业外文翻译---云计算论述：模式对比和研究问题的一些思路.docx

a tale of clouds: paradigm comparisons and some thoughts on research issues*lijun meithe university of hong kongpokfulam, hong kongljmeics.hku.hkcloud computing is an emerging computing paradigm. it aims to share data, calculations, and services transparentlyamong users of a massive grid. although theindustry has started selling cloud-computing products, research challenges in various areas, such as ui design, task decomposition, task distribution, and task coordination, are still unclear. therefore, we study the methods toreason and model cloud computing as a step towardidentifying fundamental research questions in this paradigm.in this paper, we compare cloud computing withservice computing and pervasive computing. both theindustry and research community have actively examined these three computing paradigms. we draw a qualitative comparison among them based on the classic model ofcomputer architecture. we finally evaluate the comparisonresults and draw up a series of research questions incloud computing for future exploration.keywords: cloud computing, paradigm comparison.1. introductioncloud computing is a paradigm that focuses on sharingdata and computations over a scalable network of nodes.examples of such nodes include end user computers, datacenters, and web services. we term such a network ofnodes as a cloud. an application based on such clouds istaken as a cloud application.this paradigm is increasingly popular in the industry,where industrial leaders such as microsoft 26, google2, and ibm 5 strongly promote the paradigm in recentyears. an early attempt to formulate cloud computingdates back to at least 1997 8. however, to our bestknowledge, the adoption and promotion of cloudcomputing has been slow until 2007 9.we observe that the history of early industrialadoptions of cloud computing share some commonmilestones with that of service computing 4. forexample, it took service computing 27 a long time (tenyears or so) to receive worldwide support from leadingcompanies like ibm, microsoft 25, bea, and oracle.similarly, it has been many years since the earlyformalization effort 8 toward cloud computing.besides, the wide adoption of a computing paradigmusually depends highly on the maturity of supportingtechnologies and industry recognitions. service computinghas become much more popular since the success ofweb services, although a web service is only one of thetechnologies to fulfill the notion of service orientation 4.similarly, the distributed computing community haspointed out that many distributed computing techniquesfor cloud computing have been mature 71011. manycompanies such as dell and ibm have begun to shipcloud computing machines 510.last but not the least, in either service computing orcloud computing, research developments lag behindindustrial adoptions. for instance, coscon, a leadinginternational container shipper, has a successful adoptionof service computing. it successfully used service-orientedarchitecture to improve the business responsibility tocustomers in 2004 3. yet, research studies in serviceorientedarchitecture from the software engineeringcommunity 19 are still inadequate.despite our survey over the internet, to our bestknowledge, there are few articles to pinpoint research* 2008 ieee. this material is presented to ensure timely disseminationof scholarly and technical work. personal use of this material ispermitted. copyright and all rights therein are retained by authors or byother copyright holders. all persons copying this information areexpected to adhere to the terms and constraints invoked by eachauthors copyright. in most cases, these works may not be repostedwithout the explicit permission of the copyright holder. permission toreprint/republish this material for advertising or promotional purposesor for creating new collective works for resale or redistribution toservers or lists, or to reuse any copyrighted component of this work inother works must be obtained from the ieee. this research is supported in part by the general research fund of theresearch grant council of hong kong (project nos. 111107, 717308,and 717506). corresponding author.2issues in cloud computing. this would slow down thenext research advances. we will alleviate this problem inthe present paper.in this paper, we use the classic computer architecturemodel 15 to provide a qualitative comparison frameworkto compare cloud computing with pervasive computingand service computing. the qualitative comparisonframework includes three features: input-output (i/o),storage, and calculation. for each feature, we draw thecomparison using multiple characteristics. through suchcomparisons, we identify the connections between cloudcomputing and the other two computing paradigms fromthe perspective of software engineering. based on theconnections, we draw up a few research issues and discussthem in the paper to promote future exploration.the main contribution of the paper is twofold: (i) toour best knowledge, we provide the first qualitativecomparison on cloud computing, service computing, andpervasive computing. (ii) we present a series of researchissues in cloud computing on top of the comparisonframework. these issues promote future explorations.the rest of the paper is organized as follows: section 2presents the preliminaries of cloud computing, servicecomputing, and pervasive computing. section 3 introducesour qualitative framework to compare the abovethree computing paradigms and present our efforts toidentify research issues in cloud computing. finally, wereview related work in section 4 and draw a conclusion insection 5.2. preliminariesthis section reviews the preliminaries of cloudcomputing, service computing, and pervasive computing.2.1. cloud computingas we have introduced in section 1, a computing cloudis a massive network of nodes. thus, scalability should bea quality feature of the computing cloud. it has at leasttwo dimensions, namely horizontal cloud scalability andvertical cloud scalability (adapted from 9). horizontal cloud scalability is the ability to connectand integrate multiple clouds to work as one logicalcloud. for instance, a cloud providing calculationservices (calculation cloud) can access a cloudproviding storage services (storage cloud) to keepintermediate results. two calculation clouds can alsointegrate into a larger calculation cloud. vertical cloud scalability is the ability to improve thecapacity of a cloud by enhancing individual existingnodes in the cloud (such as providing a server withmore physical memory) or improving the bandwidththat connects two nodes. in addition, to meet increasingmarket demand, a node can be gradually upgraded froma single power machine to a data center.scalability should be transparent to users. for instance,users may store their data in the cloud without the need toknow where it keeps the data or how it accesses the data.for simplicity, we will refer to horizontal and verticalcloud scalability, respectively, as horizontal scalabilityand vertical scalability in this paper.2.2. service computingservice computing (or service-oriented computing) isan emerging paradigm to model, create, operate, andmanage business services. in this paradigm, servicespublish themselves in public registries, discover peerservices, and bind to the latter services to form servicecompositions using standardized protocols 6. to create aservice composition, engineers may use a specification,such as ws-bpel 30, to model the collaborative needin workflows. to carry out individual workflow steps,software developers may use web services, the mostpopular way to fulfill service-oriented architecture in theindustry. a set of service-oriented applications over theweb services thus creates a network of services.service service registryregister serviceto registrydiscover servicefrom a registrybind service associate servicefigure 1. service-oriented network 18.we briefly describe a service-oriented network 18 tofacilitate the comparison in the rest of the paper. anelement in such a network is a service registry, serviceconsumer, or service provider. a service providerregisters itself in a service registry. a service consumerfirst discovers the service from a registry, and then bindsto the service. a service provider may register itself tomore than one registry. a registry may also associate itsregistered services to other registries, and acts as a serviceitself. such a treatment on a registry provides a genericview among elements in service-oriented modeling.2.3. pervasive computingpervasive computing (or ubiquitous computing) 2324is another emerging computing paradigm. software (oftenreferred as pervasive software) can be embedded in aconstantly changing computing environment. therefore,pervasive software users do not need to be concernedabout how to adjust the software to adapt to thesurrounding computing environment. a well-developedenvironment will enable users to use pervasive softwareeverywhere without extra effort.to understand and react to a user, applications useenvironmental features, known as contexts, extensively.sensors can capture these contexts. to allow ubiquitous3support to end users, smart sensors are placed aroundusers to preserve different information, such as thelocations, contexts, and user-relevant data.figure 2 shows a pervasive computing example.sensors, mobile phones and pdas, desktop computer, andservers are interconnected logically to form an application.suppose a nomadic user at the top left corner of figure 2moves from using a laptop to using a desktop computer.the laptop and the desktop computer both serve as uiportals to the tuple space maintained by the pervasivesoftware. the remarked information from various displayportals (such as the pdas on the right-hand part) mayneed adapting. for example, a desktop computer may beequipped with a high-definition webcam. thus, apresentation display portal may display the contents witha camera image kept in the tuple space of the applicationwhen using a laptop.figure 2. pervasive computing environment 22.3. comparison of cloud, service, andpervasive computing paradigmsthis section presents a qualitative comparison amongthe cloud, service, and pervasive computing paradigms.many researchers consider cloud computing as derivedfrom grid computing 12 and have provided manycomparisons between them 21. to identify more issuesfor cloud computing, we choose to compare it withservice computing and pervasive computing for thefollowing reasons. service computing is useful inmodeling functionality and providing flexible services.pervasive computing enables users to use softwareeverywhere and provides self-adaptive capacity to thesoftware with respect to environmental contexts. cloudcomputing needs both functionality modeling and contextsensitivity.through comparison with service computingand pervasive computing, therefore, we can gain insightson cloud computing.researchers (such as 13) have applied the notion ofvirtual computers to model various computing entities andtheir interconnections. such a treatment motivates us toanalyze the key features of software applications (orservices) of cloud computing. we thus compare cloudcomputing with service computing and pervasivecomputing from the perspective of computer architecture.in classic computer architecture 15, a computer hasthree features: input-output (i/o), storage, and calculation.the descriptions of these features are as follows:(i) the typical computer-input entities include the keyboardand mouse, and the computer-output entities include, forinstance, the monitor, printer, and speaker. (ii) in thestorage feature, there are storage entities such as the harddisk (internal storage) and usb (external storage).(iii) the key entity in the calculation feature is the cpu.we then show the representative characteristics in eachfeature of the three computing paradigms from theperspective of software engineering. our understanding ofservice computing and pervasive computing is mainlybased on our software engineering research 171819in these two paradigms. our understanding of cloudcomputing is mainly based on our survey over the internet.we summarize the comparison results in table 1. thekey findings are as follows. we note that at least threenotable likenesses of cloud computing from table 1: the i/o feature of cloud computing resembles that ofservice computing. the storage feature of cloud computing is closer tothat of pervasive computing than service computing. the calculation features of the three computingparadigms are similar.table 1. comparisons in the framework of theclassical model of computer architecturemodeldimension general characteristicsi/o user requests and cloud responsesstorage stored in the clouds collectivelycloudcomputingcalculation both intra-cloud calculationsand inter-cloud calculationsi/o service requests and service responsesstorage stored in specific service hosts servicecomputingcalculation performed by individual servicecompositionsi/o situation detections and setupstorage stored in the tuple space of theapplicationpervasivecomputingcalculationmainly performed by the entitiesembedded or connected to thesurrounding environmentswe thus use the above-identified likenesses to extractthe main properties of service computing and pervasivecomputing to study cloud computing. although the threeparadigms are similar at a high-level, they still showdifferences in the details, as we will present below.tables 2, 3, and 4 show the comparisons of the keyproperties (that is, subfeatures) in the i/o, storage, andcalculation features, respectively. owing to the page limit,we will leave the comparisons of other properties infurther publications. we only pick the key points of themain properties for discussions in this paper.in the sequel, we will discuss the research questionsindexed in tables 2, 3, and 4.4table 2. comparisons in the i/o featurei/o attributes description(research question index)interface cloud interface (q1, q2)(not yet formally defined)data type cloud data type (q2)(not yet formally defined)cloudcomputingsynchronizationsynchronous or asynchronous i/ocommunication? (not yet formallydefined) (q2)interface service interfacedata type xml data which can be transferredusing certain protocols (e.g., soap)servicecomputingsynchronization providing both synchronous andasynchronous i/o communicationsinterfaceinterfaces with various devices inthe environments (e.g., pdas,mobile phones, and laptops)data type various data types (e.g., xml,wap, gprs, and bluetooth)pervasivecomputingsynchronization providing both synchronous andasynchronous i/o communicationstable 3. comparisons in the storage featurestorageattributesdescription(research question index)locationencapsulated in clouds. no explicitdistinction between local and remotestorage entities (q2)scale the scale of intra-cloud storage and theinter-cloud storage (q1 )cloudcomputingaccess through cloud access (q2)locationencapsulated within individual services.online storage is not the focus of servicecomputingscale depending on the storage scales ofindividual service hostsservicecomputingaccess service requestslocation explicit storage in the surroundingenvironmentsscaledepending on the storage scales in theenvironment or inter-connected to theenvironmentpervasivecomputingaccess through context communicationsq1. how do computing entities dynamically pluginto a computing cloud?in service computing, service providers dynamicallyregister their services into the public service registries.service consumers discover services from the registriesand dynamically bind or unbind themselves to theseservices 18. in pervasive computing, a mobile entity canmove from one place to another and embed into differentenvironments 17.similarly, in cloud computing, computing entitiesshould be able to plug into a cloud dynamically. forexample, when a large cluster of computer workstationsand business services are attached to a cloud, the availabilityof computing entities in the cloud may changeradically. how can a cloud application be entity-aware toplug-in heterogonous computing entities (which may alsobe different in modeling) with respect to variousapplication parts? techniques to alleviate this problemwill strengthen the vertical scalability of applications.table 4. comparisons in the calculation featurecalculationattributesdescription(research question index)locationencapsulated in clouds. no explicitdistinction between local and remotecalculation entities. (q1, q2)context environment of individual entities inclouds. unclear location of contexts. (q3)cloudcomputinggranularity the entire cloud. (q4)locationencapsulated within individual services.discovery of services via explicit servicepublishing, discovering, and bindingmechanisms.contextenvironment of individual services andservice compositions. contexts are kept inservices.servicecomputinggranularity usually, a small subset of all availableservices in a son.location explicit calculation discoveries viacontext communications.context environment of individual computingentities. context