計算機專業(yè)外文翻譯---云計算論述模式對比和研究問題的一些思路

1、<p>　　A Tale of Clouds: Paradigm Comparisons and Some Thoughts on Research Issues*?</p><p><b>　　Lijun Mei</b></p><p>　　The University of Hong Kong</p><p>　　Pokfulam

2、, Hong Kong</p><p>　　ljmei@cs.hku.hk</p><p>　　Cloud computing is an emerging computing paradigm. It aims to share data, calculations, and services transparentlyamong users of a massive grid. Alt

3、hough 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 stud

4、y the methods toreason and model cloud computing as a step towardidentifying fundamental research questions in this paradigm.</p><p>　　In this paper, we compare cloud computing withservice computing and perv

5、asive computing. Both the</p><p>　　industry and research community have actively examined these three computing paradigms. We draw a qualitative comparison among them based on the classic model of</p>

6、<p>　　computer architecture. We finally evaluate the comparison</p><p>　　results and draw up a series of research questions in</p><p>　　cloud computing for future exploration.</p>&

7、lt;p>　　Keywords: cloud computing, paradigm comparison.</p><p>　　1. Introduction</p><p>　　Cloud computing is a paradigm that focuses on sharing</p><p>　　data and computations over

8、 a scalable network of nodes.</p><p>　　Examples of such nodes include end user computers, data</p><p>　　centers, and Web Services. We term such a network of</p><p>　　nodes as a clou

9、d. An application based on such clouds is</p><p>　　taken as a cloud application.</p><p>　　This paradigm is increasingly popular in the industry,</p><p>　　where industrial leaders su

10、ch as Microsoft [26], Google</p><p>　　[2], and IBM [5] strongly promote the paradigm in recent</p><p>　　years. An early attempt to formulate cloud computing</p><p>　　dates back to a

11、t least 1997 [8]. However, to our best</p><p>　　knowledge, the adoption and promotion of cloud</p><p>　　computing has been slow until 2007 [9].</p><p>　　We observe that the history

12、of early industrial</p><p>　　adoptions of cloud computing share some common</p><p>　　milestones with that of service computing [4]. For</p><p>　　example, it took service computing [

13、27] a long time (ten</p><p>　　years or so) to receive worldwide support from leading</p><p>　　companies like IBM, Microsoft [25], BEA, and Oracle.</p><p>　　Similarly, it has been ma

14、ny years since the early</p><p>　　formalization effort [8] toward cloud computing.</p><p>　　Besides, the wide adoption of a computing paradigm</p><p>　　usually depends highly on the

15、 maturity of supporting</p><p>　　technologies and industry recognitions. Service computing</p><p>　　has become much more popular since the success of</p><p>　　Web services, although

16、 a Web service is only one of the</p><p>　　technologies to fulfill the notion of service orientation [4].</p><p>　　Similarly, the distributed computing community has</p><p>　　pointe

17、d out that many distributed computing techniques</p><p>　　for cloud computing have been mature [7][10][11]. Many</p><p>　　companies such as Dell and IBM have begun to ship</p><p>　　

18、cloud computing machines [5][10].</p><p>　　Last but not the least, in either service computing or</p><p>　　cloud computing, research developments lag behind</p><p>　　industrial adop

19、tions. For instance, COSCON, a leading</p><p>　　international container shipper, has a successful adoption</p><p>　　of service computing. It successfully used service-oriented</p><p&g

20、t;　　architecture to improve the business responsibility to</p><p>　　customers in 2004 [3]. Yet, research studies in serviceoriented</p><p>　　architecture from the software engineering</p>

21、<p>　　community [19] are still inadequate.</p><p>　　Despite our survey over the Internet, to our best</p><p>　　knowledge, there are few articles to pinpoint research</p><p>　　

22、* © 2008 IEEE. This material is presented to ensure timely dissemination</p><p>　　of scholarly and technical work. Personal use of this material is</p><p>　　permitted. Copyright and all rig

23、hts therein are retained by authors or by</p><p>　　other copyright holders. All persons copying this information are</p><p>　　expected to adhere to the terms and constraints invoked by each</

24、p><p>　　author’s copyright. In most cases, these works may not be reposted</p><p>　　without the explicit permission of the copyright holder. Permission to</p><p>　　reprint/republish th

25、is material for advertising or promotional purposes</p><p>　　or for creating new collective works for resale or redistribution to</p><p>　　servers or lists, or to reuse any copyrighted component

26、 of this work in</p><p>　　other works must be obtained from the IEEE.</p><p>　　? This research is supported in part by the General Research Fund of the</p><p>　　Research Grant Counc

27、il of Hong Kong (project nos. 111107, 717308,</p><p>　　and 717506).</p><p>　　? Corresponding author.</p><p><b>　　2</b></p><p>　　issues in cloud computing. T

28、his would slow down the</p><p>　　next research advances. We will alleviate this problem in</p><p>　　the present paper.</p><p>　　In this paper, we use the classic computer architectu

29、re</p><p>　　model [15] to provide a qualitative comparison framework</p><p>　　to compare cloud computing with pervasive computing</p><p>　　and service computing. The qualitative com

30、parison</p><p>　　framework includes three features: input-output (I/O),</p><p>　　storage, and calculation. For each feature, we draw the</p><p>　　comparison using multiple character

31、istics. Through such</p><p>　　comparisons, we identify the connections between cloud</p><p>　　computing and the other two computing paradigms from</p><p>　　the perspective of softwa

32、re engineering. Based on the</p><p>　　connections, we draw up a few research issues and discuss</p><p>　　them in the paper to promote future exploration.</p><p>　　The main contribut

33、ion of the paper is twofold: (i) To</p><p>　　our best knowledge, we provide the first qualitative</p><p>　　comparison on cloud computing, service computing, and</p><p>　　pervasive c

34、omputing. (ii) We present a series of research</p><p>　　issues in cloud computing on top of the comparison</p><p>　　framework. These issues promote future explorations.</p><p>　　The

35、 rest of the paper is organized as follows: Section 2</p><p>　　presents the preliminaries of cloud computing, service</p><p>　　computing, and pervasive computing. Section 3 introduces</p>

36、<p>　　our qualitative framework to compare the above</p><p>　　three computing paradigms and present our efforts to</p><p>　　identify research issues in cloud computing. Finally, we</p&g

37、t;<p>　　review related work in Section 4 and draw a conclusion in</p><p>　　Section 5.</p><p>　　2. Preliminaries</p><p>　　This section reviews the preliminaries of cloud</p

38、><p>　　computing, service computing, and pervasive computing.</p><p>　　2.1. Cloud computing</p><p>　　As we have introduced in Section 1, a computing cloud</p><p>　　is a ma

39、ssive network of nodes. Thus, scalability should be</p><p>　　a quality feature of the computing cloud. It has at least</p><p>　　two dimensions, namely horizontal cloud scalability and</p>

40、<p>　　vertical cloud scalability (adapted from [9]).</p><p>　　􀁺?Horizontal cloud scalability is the ability to connect</p><p>　　and integrate multiple clouds to work as one logica

41、l</p><p>　　cloud. For instance, a cloud providing calculation</p><p>　　services (calculation cloud) can access a cloud</p><p>　　providing storage services (storage cloud) to keep<

42、;/p><p>　　intermediate results. Two calculation clouds can also</p><p>　　integrate into a larger calculation cloud.</p><p>　　􀁺?Vertical cloud scalability is the ability to imp

43、rove the</p><p>　　capacity of a cloud by enhancing individual existing</p><p>　　nodes in the cloud (such as providing a server with</p><p>　　more physical memory) or improving the b

44、andwidth</p><p>　　that connects two nodes. In addition, to meet increasing</p><p>　　market demand, a node can be gradually upgraded from</p><p>　　a single power machine to a data ce

45、nter.</p><p>　　Scalability should be transparent to users. For instance,</p><p>　　users may store their data in the cloud without the need to</p><p>　　know where it keeps the data o

46、r how it accesses the data.</p><p>　　For simplicity, we will refer to horizontal and vertical</p><p>　　cloud scalability, respectively, as horizontal scalability</p><p>　　and vertic

47、al scalability in this paper.</p><p>　　2.2. Service computing</p><p>　　Service computing (or service-oriented computing) is</p><p>　　an emerging paradigm to model, create, operate,

48、and</p><p>　　manage business services. In this paradigm, services</p><p>　　publish themselves in public registries, discover peer</p><p>　　services, and bind to the latter services

49、to form service</p><p>　　compositions using standardized protocols [6]. To create a</p><p>　　service composition, engineers may use a specification,</p><p>　　such as WS-BPEL [30], t

50、o model the collaborative need</p><p>　　in workflows. To carry out individual workflow steps,</p><p>　　software developers may use Web services, the most</p><p>　　popular way to ful

51、fill service-oriented architecture in the</p><p>　　industry. A set of service-oriented applications over the</p><p>　　Web services thus creates a network of services.</p><p>　　Servi

52、ce Service Registry</p><p>　　Register Service</p><p>　　to Registry</p><p>　　Discover Service</p><p>　　from a registry</p><p>　　Bind Service Associate Servi

53、ce</p><p>　　Figure 1. Service-oriented network [18].</p><p>　　We briefly describe a service-oriented network [18] to</p><p>　　facilitate the comparison in the rest of the paper. An&

54、lt;/p><p>　　element in such a network is a service registry, service</p><p>　　consumer, or service provider. A service provider</p><p>　　registers itself in a service registry. A servi

55、ce consumer</p><p>　　first discovers the service from a registry, and then binds</p><p>　　to the service. A service provider may register itself to</p><p>　　more than one registry.

56、A registry may also associate its</p><p>　　registered services to other registries, and acts as a service</p><p>　　itself. Such a treatment on a registry provides a generic</p><p>　

57、　view among elements in service-oriented modeling.</p><p>　　2.3. Pervasive computing</p><p>　　Pervasive computing (or ubiquitous computing) [23][24]</p><p>　　is another emerging com

58、puting paradigm. Software (often</p><p>　　referred as pervasive software) can be embedded in a</p><p>　　constantly changing computing environment. Therefore,</p><p>　　pervasive soft

59、ware users do not need to be concerned</p><p>　　about how to adjust the software to adapt to the</p><p>　　surrounding computing environment. A well-developed</p><p>　　environment wi

60、ll enable users to use pervasive software</p><p>　　everywhere without extra effort.</p><p>　　To understand and react to a user, applications use</p><p>　　environmental features, kno

61、wn as contexts, extensively.</p><p>　　Sensors can capture these contexts. To allow ubiquitous</p><p><b>　　3</b></p><p>　　support to end users, smart sensors are placed a

62、round</p><p>　　users to preserve different information, such as the</p><p>　　locations, contexts, and user-relevant data.</p><p>　　Figure 2 shows a pervasive computing example.</

63、p><p>　　Sensors, mobile phones and PDAs, desktop computer, and</p><p>　　servers are interconnected logically to form an application.</p><p>　　Suppose a nomadic user at the top left cor

64、ner of Figure 2</p><p>　　moves from using a laptop to using a desktop computer.</p><p>　　The laptop and the desktop computer both serve as UI</p><p>　　portals to the tuple space mai

65、ntained by the pervasive</p><p>　　software. The remarked information from various display</p><p>　　portals (such as the PDAs on the right-hand part) may</p><p>　　need adapting. For

66、example, a desktop computer may be</p><p>　　equipped with a high-definition webcam. Thus, a</p><p>　　presentation display portal may display the contents with</p><p>　　a camera imag

67、e kept in the tuple space of the application</p><p>　　when using a laptop.</p><p>　　Figure 2. Pervasive computing environment [22].</p><p>　　3. Comparison of cloud, service, and<

68、/p><p>　　pervasive computing paradigms</p><p>　　This section presents a qualitative comparison among</p><p>　　the cloud, service, and pervasive computing paradigms.</p><p>

69、;　　Many researchers consider cloud computing as derived</p><p>　　from grid computing [12] and have provided many</p><p>　　comparisons between them [21]. To identify more issues</p><p&

70、gt;　　for cloud computing, we choose to compare it with</p><p>　　service computing and pervasive computing for the</p><p>　　following reasons. Service computing is useful in</p><p>　

71、　modeling functionality and providing flexible services.</p><p>　　Pervasive computing enables users to use software</p><p>　　everywhere and provides self-adaptive capacity to the</p><

72、p>　　software with respect to environmental contexts. Cloud</p><p>　　computing needs both functionality modeling and contextsensitivity.</p><p>　　Through comparison with service computing</

73、p><p>　　and pervasive computing, therefore, we can gain insights</p><p>　　on cloud computing.</p><p>　　Researchers (such as [13]) have applied the notion of</p><p>　　virtu

74、al computers to model various computing entities and</p><p>　　their interconnections. Such a treatment motivates us to</p><p>　　analyze the key features of software applications (or</p>&

75、lt;p>　　services) of cloud computing. We thus compare cloud</p><p>　　computing with service computing and pervasive</p><p>　　computing from the perspective of computer architecture.</p>

76、<p>　　In classic computer architecture [15], a computer has</p><p>　　three features: Input-Output (I/O), Storage, and Calculation.</p><p>　　The descriptions of these features are as follow

77、s:</p><p>　　(i) The typical computer-Input entities include the keyboard</p><p>　　and mouse, and the computer-Output entities include, for</p><p>　　instance, the monitor, printer, a

78、nd speaker. (ii) In the</p><p>　　Storage feature, there are storage entities such as the hard</p><p>　　disk (internal storage) and USB (external storage).</p><p>　　(iii) The key ent

79、ity in the Calculation feature is the CPU.</p><p>　　We then show the representative characteristics in each</p><p>　　feature of the three computing paradigms from the</p><p>　　persp

80、ective of software engineering. Our understanding of</p><p>　　service computing and pervasive computing is mainly</p><p>　　based on our software engineering research [17][18][19]</p><

81、p>　　in these two paradigms. Our understanding of cloud</p><p>　　computing is mainly based on our survey over the Internet.</p><p>　　We summarize the comparison results in Table 1. The</p&g

82、t;<p>　　key findings are as follows. We note that at least three</p><p>　　notable likenesses of cloud computing from Table 1:</p><p>　　􀁺?The I/O feature of cloud computing resemb

83、les that of</p><p>　　service computing.</p><p>　　􀁺?The storage feature of cloud computing is closer to</p><p>　　that of pervasive computing than service computing.</p>

84、;<p>　　􀁺?The calculation features of the three computing</p><p>　　paradigms are similar.</p><p>　　Table 1. Comparisons in the framework of the</p><p>　　classical mod

85、el of computer architecture</p><p><b>　　Model</b></p><p>　　Dimension General Characteristics</p><p>　　I/O User requests and cloud responses</p><p>　　Storage

86、 Stored in the clouds collectively</p><p><b>　　Cloud</b></p><p><b>　　Computing</b></p><p>　　Calculation Both intra-cloud calculations</p><p>　　a

87、nd inter-cloud calculations</p><p>　　I/O Service requests and service responses</p><p>　　Storage Stored in specific service hosts Service</p><p><b>　　Computing</b></p

88、><p>　　Calculation Performed by individual service</p><p>　　compositions</p><p>　　I/O Situation detections and setup</p><p>　　Storage Stored in the tuple space of the</

89、p><p>　　application</p><p><b>　　Pervasive</b></p><p><b>　　Computing</b></p><p>　　Calculation</p><p>　　Mainly performed by the entities

90、</p><p>　　embedded or connected to the</p><p>　　surrounding environments</p><p>　　We thus use the above-identified likenesses to extract</p><p>　　the main properties of

91、 service computing and pervasive</p><p>　　computing to study cloud computing. Although the three</p><p>　　paradigms are similar at a high-level, they still show</p><p>　　differences

92、 in the details, as we will present below.</p><p>　　Tables 2, 3, and 4 show the comparisons of the key</p><p>　　properties (that is, subfeatures) in the I/O, storage, and</p><p>　　c

93、alculation features, respectively. Owing to the page limit,</p><p>　　we will leave the comparisons of other properties in</p><p>　　further publications. We only pick the key points of the</p&

94、gt;<p>　　main properties for discussions in this paper.</p><p>　　In the sequel, we will discuss the research questions</p><p>　　indexed in Tables 2, 3, and 4.</p><p><b>

95、;　　4</b></p><p>　　Table 2. Comparisons in the I/O feature</p><p>　　I/O Attributes Description</p><p>　　(Research Question Index)</p><p>　　Interface Cloud interfac

96、e (Q1, Q2)</p><p>　　(not yet formally defined)</p><p>　　Data Type Cloud data type (Q2)</p><p>　　(not yet formally defined)</p><p><b>　　Cloud</b></p>

97、<p><b>　　Computing</b></p><p>　　Synchronization</p><p>　　Synchronous or asynchronous I/O</p><p>　　communication? (Not yet formally</p><p>　　defined) (Q

98、2)</p><p>　　Interface Service interface</p><p>　　Data Type XML data which can be transferred</p><p>　　using certain protocols (e.g., SOAP)</p><p><b>　　Service<

99、/b></p><p><b>　　Computing</b></p><p>　　Synchronization Providing both synchronous and</p><p>　　asynchronous I/O communications</p><p><b>　　Interfac

100、e</b></p><p>　　Interfaces with various devices in</p><p>　　the environments (e.g., PDAs,</p><p>　　mobile phones, and laptops)</p><p>　　Data Type Various data type

101、s (e.g., XML,</p><p>　　WAP, GPRS, and Bluetooth)</p><p><b>　　Pervasive</b></p><p><b>　　Computing</b></p><p>　　Synchronization Providing both syn

102、chronous and</p><p>　　asynchronous I/O communications</p><p>　　Table 3. Comparisons in the Storage feature</p><p><b>　　Storage</b></p><p>　　Attributes</p

103、><p>　　Description</p><p>　　(Research Question Index)</p><p><b>　　Location</b></p><p>　　Encapsulated in clouds. No explicit</p><p>　　distinction b

104、etween local and remote</p><p>　　storage entities (Q2)</p><p>　　Scale The scale of intra-cloud storage and the</p><p>　　inter-cloud storage (Q1 )</p><p><b>　　Clou

105、d</b></p><p><b>　　Computing</b></p><p>　　Access Through cloud access (Q2)</p><p><b>　　Location</b></p><p>　　Encapsulated within individual

106、 services.</p><p>　　Online storage is not the focus of service</p><p><b>　　computing</b></p><p>　　Scale Depending on the storage scales of</p><p>　　individu

107、al service hosts</p><p><b>　　Service</b></p><p><b>　　Computing</b></p><p>　　Access Service requests</p><p>　　Location Explicit storage in the su

108、rrounding</p><p>　　environments</p><p><b>　　Scale</b></p><p>　　Depending on the storage scales in the</p><p>　　environment or inter-connected to the</p&g

109、t;<p>　　environment</p><p><b>　　Pervasive</b></p><p><b>　　Computing</b></p><p>　　Access Through context communications</p><p>　　Q1. How d

110、o computing entities dynamically plug</p><p>　　into a computing cloud?</p><p>　　In service computing, service providers dynamically</p><p>　　register their services into the public

111、service registries.</p><p>　　Service consumers discover services from the registries</p><p>　　and dynamically bind or unbind themselves to these</p><p>　　services [18]. In pervasive

112、 computing, a mobile entity can</p><p>　　move from one place to another and embed into different</p><p>　　environments [17].</p><p>　　Similarly, in cloud computing, computing entiti

113、es</p><p>　　should be able to plug into a cloud dynamically. For</p><p>　　example, when a large cluster of computer workstations</p><p>　　and business services are attached to a clo

114、ud, the availability</p><p>　　of computing entities in the cloud may change</p><p>　　radically. How can a cloud application be entity-aware to</p><p>　　plug-in heterogonous computin