Cloud Computing: Concept, Model, and Key Technologies

  Cloud computing concepts, technologies, and terminologies are continually emerging and being discussed. In particular, interest is growing in the implementation of cloud computing for enterprises. Among industry professionals, there is a general level of understanding of the benefits cloud computing might bring. However, cloud computing is a new concept with a broad definition, and some in the marketing sector have espoused it as an omnipotent technology that includes everything, enables everything, and is ubiquitous. A certain degree of hype surrounds cloud computing, making the concept itself seem like real clouds: scudding and illusory.  Unfortunately, marketing hype gives rise to misunderstanding about cloud computing and steers the cloud computing market blindly. This is detrimental to the growth of cloud computing in China. Many introductory articles have been written on cloud computing in China, but there are few papers that analyze its trends rationally. Therefore, a comprehensive clarification of the origin and definition of cloud computing is necessary. This paper seeks to analyze early infrastructure, technology models, and service forms of cloud computing and to summarize advantages and disadvantages so that a breakthrough direction can be ascertained for future cloud computing development.

  It should be noted that current discussions on cloud computing in China are focused on the concept, technology, and model in its early days. Cloud computing was initially defined as a dynamic, scalable computing model for offering virtualized IT resources and applications over the Internet. Consumers are not required to have technical expertise, nor are they directly involved in investment, establishment, maintenance, or control in cloud computing. They use cloud services on demand, and pay only for what is used. The essential purpose of early cloud computing technology was to enable consumers to share a common physical resource pool by using virtualization technology. However, virtualization is not all of cloud computing. IT resource components in early cloud computing systems include servers, storage, bandwidth, networks, and security. Data centers are central in supporting and supplying resources and capabilities. Early cloud computing practice integrated virtualization, grid computing, and distributed computing technologies. It represented not only the innovative combining of emerging technologies, but also innovation in service and commercial models. Early cloud computing has influenced development trends in IT technology and IT infrastructure, as well as in IT   commercial models, application development, and service deployment and operation.

  Well-known and widely introduced, early cloud technology has been commercially deployed, and has achieved some successes. However, the philosophy of early cloud computing still has many limitations. In this paper, these limitations are traced to their source, and the next stage of development is discussed from an industrial perspective. For the majority of enterprises, data center transformation is the foremost task of cloud computing. Data centre transformation will popularize cloud computing technology and enable widespread application in businesses, institutions, and organizations[1-4]. The early cloud computing concept, technology, and model will be expanded in an all-round way. Services based on traditional IT resources and applications will continue to be provided, but support will also be provided for resources and applications arising from the integration of IT, communications, TV, mobile, and Internet of Things technologies. There is little doubt that cloud computing, as a new technology and new operational model, will profoundly affect IT industrial development in the future.

1 Concept of Cloud Computing
  Computer application models have generally developed in the following stages: large computer-based centralized architecture (Data Center 1.0); PC-based client/server distributed computing architecture (Data Center 2.0); virtualization-based Service-Oriented Architecture (SOA), and new-type architecture based on Web2.0 application characteristics (Data Center 3.0). In this context, the architecture and implementation characteristics of cloud computing have developed.

  Cloud computing is a popular term, but it is still difficult to conclusively define. Understanding the term as it is most commonly used is necessary for further discussion.
"Computing" generally refers to computing application; that is, any IT application in industry or in the market.  Because network technologies are being converged, all applications in information, communication, and video are integrated on a unified platform. Likewise, computing in cloud computing refers to any integrated ICT application. It is a straightforward concept. The key characteristic of cloud computing is not "computing" but "cloud."
The term "cloud" was born quite accidentally. In the early days of the Internet, a cloud was usually drawn to represent the Internet. The inner workings and complex mechanisms of networks were abstracted to simplify discussions of new technologies. The pervasive application of the Internet has fundamentally changed ICT application architectures. In this context, the term "cloud computing" is naturally used to represent new network computing characteristics and technology trends. This is the reason the term has become so popular.

    Internet technologies become the basis of ICT applications, and ICT itself requires reconsidering and redesigning in response to new Internet application demands. As the Internet changes, ICT application platforms need to be morphed, and more innovative ICT application implementation and commercial models emerge. These changes may have such a great effect that people may perceive the new characteristics and phenomena of ICT from different perspectives. Despite having many definitions, cloud computing is essentially used for network implementation of ICT applications and services. Cloud computing here is clearly and strictly defined: Based on virtualized resources of integrated infrastructure, cloud computing is the ICT service delivery model for scalable ICT applications over IP networks.

    The essence of cloud computing is applications over network; applications are the joint products of IP and IT technologies. The technologies and goals of clouds are continuously evolving. When Web technology emerged, it also had application characteristics of cloud computing and the prototype of a unified interface. As virtualization technology developed on the server application platform and Web-unified interfaces were launched, virtualization and Web were integrated for unified cloud computing over an integrated architecture.

2 Implementation Models of Cloud Computing
    If clouds are essentially service implementation models, what do new cloud computing service models look like?

    Amazon, Google, and Saleforces.com are well-known contributors to early cloud computing. They have successfully provided cloud services with distinguishable characteristics and using different models.

    In Infrastructure as a Service (IaaS), IT infrastructure is used as a service platform to provide cloud services according to duration of use and amount of resources. IaaS consumers can be individuals or enterprises, institutions or organizations. Amazon Elastic Compute Cloud (EC2) is the first IaaS in the industry. Computing, storage, network, and security resources are rented to consumers, who use them with operating systems and application software.

    In Platform as a Service (PaaS), application development environments are used as service platforms in order to provide development interfaces and tools for users to create new applications. Services within clouds are implemented by Internet and service providers. PaaS may use both internal and external IaaS. Google’s AppEngine software environment provides application developers with PaaS, and developers use AppEngine to develop new applications.
Software as a Service (SaaS) refers to software based on IaaS or PaaS. As opposed to traditional software sets, SaaS implements services by Internet applications. SaaS is available in either internal or external IaaS/PaaS. Saleforces.com is perhaps the most well-known SaaS operator, offering Enterprise Resource Planning (ERP) application services. SaaS application programs are centralized and managed by SaaS operators; so users are saved from installation, maintenance, upgrade, and management.

    A cloud computing system consists of three layers: SaaS on the top, PaaS in the middle, and IaaS on the bottom. Below IaaS are the basic technologies of building block of cloud computing.

    Basic characteristics of cloud computing, as observed from operational cases, are:

    (1) Dynamical Scalability
    Cloud technologies allow users to dynamically scale IT resources up and down according to requirements. Applications are operated on virtual platforms, and no pre-reserved fixed resources are locked. The capacity of a cloud service can be dynamically adjusted over certain periods, for special applications, and in response to a changing number of users.

    (2) Very Large Scale Virtualization
    Cloud service requirements and use are independent from specific physical resources. IT applications and services are operated on virtual platforms. Cloud computing enables users to access application services with a Web browser of any kind and an Internet connection. All required resources come from very large cloud platforms.

    (3) High Availability
    Cloud platforms guarantee high availability of services by replicating data for fault tolerance and using interchangeable homogeneous computing nodes. When a fault occurs at a physical resource point, applications on this point will move to other physical resources to keep working. The user knows nothing about this movement. Therefore, cloud computing has a higher availability than other computing means.

    (4) Use as Needed and Pay-per-use
    Clouds form a huge resource pool. Consumers buy cloud services according to their needs, and pay only for what they actually use—just like public utilities. In the short term and in the long term, commercial models will support pay-for-capacity used.

    (5) Multiplexed Resources and Lower Cost
    Statistical multiplexing technology is used in cloud computing resources. This greatly improves the utilization of physical resources, and sharply lowers the cost of cloud services.
Early cloud service models share a common characteristic: they use public cloud architecture to offer stand-alone cloud services. A cloud is built and operated by the same operator, and services are made publicly available through the Internet. This cloud is called public cloud. Stand-alone refers to clouds with single vertical applications, such as search application and IT resource application.

    The concept of cloud computing was initially proposed for using the Internet to combine cheap computing resources scattered in different locations into a huge virtualized resource pool where stand-alone services  could be offered. Cloud computing makes full use of idle resources to complete a great amount of computing. Moreover, it flexibly schedules resource capacity, and responds quickly to changes in resource demand.  Cloud computing improves availability and utilization of computing resources, flexibility and scalability of applications, and service manageability. The goal of green and energy efficient computing will also eventually be fulfilled with lower operational costs.

    Although the advantages of virtualization are evident in early cloud computing, and successes have been achieved market operation, cloud computing still has some limitations:
?There is no quality service (QoS) guarantee or controllable security mechanism for services, both of which are critically important to most IT applications.

• Implementation of cloud services is based on special private protocols. Therefore, there is a risk that services can be locked by operators.
• Stand-alone clouds limit diversity of cloud service types; that is, not all enterprise IT applications can be implemented in public clouds.

    The reason for these limitations is that the public cloud model is built over the Internet, having nothing to do with network infrastructure, and cloud services are implemented using a best-effort model. Since virtualization for early cloud computing is based on private protocols, it is almost impossible for PaaS and SaaS to support cross-operator workload or workload transfer. This significantly restrains the expansion of cloud applications and services. Therefore, early cloud computing technology cannot be used for extensive ICT enterprise services, line services, or public user services, but only for specific stand-alone IT cloud services.

    A Rich Media Cloud is a platform supporting multimedia applications such as voice, data, information, video, mobile, and Internet of Things. A user can have all the cloud resources, and be the sole user of the platform. This kind of platform is called Internal Cloud. A Private Cloud can be used and controlled by the user—it can be a cloud owned by the user himself, or a rented part of a public cloud. It may even be a combination of the two.

    As a matter of urgency, the focus over the next five years will shift to internal and private clouds required by enterprises, organizations, institutions, and operators. This shift will mark a golden period for cloud computing, with a larger market and extended range of services. Currently, efforts should be focused on grasping the advantages of early clouds; integrating virtualization technologies for computing, application, and network; and offering new cloud computing services that are reliable, have controllable security, and are manageable. Open standards and industry alliances should be promoted so that rich media cloud systems and their operational models can be provided, and data centers and ICT applications transformed.

3 Evolution of Enterprise Data Center Architecture
    Both public and private clouds require the support of powerful data centers and IP networks. The emphasis of cloud computing development in the next stage will lie on IT and communications applications of enterprises, industries, institutions, and operators. Therefore, further research is needed into requirements of enterprise IT applications, challenges in data centre evolution, and key technologies in the transition from enterprise data centers to internal clouds.

3.1 Challenges Facing Traditional Enterprise IT Applications
    Data centers are platforms of traditional enterprise IT applications. They have stack architecture, including IT resources and isolated IT applications. However, as the number of enterprise IT applications grow, traditional data center architecture will no longer be capable of meeting market demands. In past years, the total number of resources—including servers and storage devices—increased by 40%-70% each year, while the average utilization of resources was only 10%-25%. An increase in the number of physical resources leads to increased electricity and cooling costs as a proportion of overall data center costs. This proportion has reached up to 25%-30% in recent years. A large number of physical resources also makes data center deployment more complicated, and increases the risk of data center faults due to human error. Up to 54% of data centre faults involve human error, and this can blow out maintenance costs too. ICT applications are continually emerging, Web2.0 applications are being quickly implemented and deployed, and SOA is developing quickly. Therefore, transition from enterprise data centers to internal clouds will become imperative.

3.2 Implementation of Internal Cloud Architecture
    Future cloud computing architecture for enterprise data centers will be designed to improve utilization of physical resources, to enable data centre sharing between multiple users, and to enable dynamic sharing of physical resources by multiuser applications with security isolation. By retaining the advantages of early cloud computing and legacy data centers, future data centers will be safe, reliable, controllable, and manageable. These new data centers will be internal clouds of enterprises, implemented by integration, virtualization, and automation.

    (1) Consolidation
    In future data center architecture, IT networks should be consolidated.

    Bottom-layer functions required by applications should be integrated with bottom-layer facilitates. The service layer should be specialized to efficiency of industrial IT applications rather than to the basic functions of security, reliability, and availability. Evolving the application silo structure towards network-centralized platform architecture is also an important aspect of consolidating data centers. All IT resources connected to application servers in the silo structure should be transitioned into networks. Networks connect various IT resources and basic functions, and become a platform for interchanging virtualized data resources. This platform provides logic services to physical resources, and dynamically deploys services to meet application needs. On such a platform, all types of applications can share computing services offered by virtualized resources.

    Generally speaking, an enterprise considers consolidating its data centers once the centers have been proven inefficient, wasteful, and energy intensive. If advanced cloud technologies can be adopted in integration design, enterprise cloud architecture would develop quickly. 

    (2) Virtualization
    Virtualization is a scheduling model of consolidated resources that is independent from physical location, physical presence, and physical status. It is a process of qualitative transition from physical resources to service forms. Virtualization plays a key role in multiplexing physical resources, reducing O&M complexity, and improving equipment utilization. It also lays a foundation for automatic resource scheduling and configuration.

    However, even though virtualization is a hot topic, most enterprises do not know how to virtualize their data centers. Consolidation of data centers is a premise for virtualization; in well integrated architecture, virtualization is not difficult. 

 
    (3) Automation
    With virtualization, intelligent systems can automatically and dynamically schedule and manage bottom-layer resources and functions. The administrator transfers application policy to the intelligent system, and the system schedules related physical resources according to optimal computing and resource configuration. In this way, limited resources are used to the maximum extent for service provision, and faults caused by the administrator are reduced to a minimum. This is an ideal resource scheduling model that conforms to the goal of cloud computing. Automatic deployment is a basic attribute of service-oriented data centers. Dynamic changing of resources, rapid scheduling, and flexible deployment can be fulfilled by automatically deploying resources according to capacity changes of data services.

3.3 Internal Cloud Models
    Internal clouds can reduce the costs of establishment and maintenance, improve efficiency in service development and deployment, and provide service-on-demand to end users. Its implementation should take into account the long-term practical experience of data centers of large-scale enterprises. IP/IT architecture should be integrated and unified, and service resources should be flexibly scheduled and shared. For diverse, safe, rich media services that have reliable quality, advanced technologies should be used in internal clouds.

    (1) Unified Fabrics 
    Unified fabrics rely on innovative technologies, represented by 10G Data Center Ethernet (DCE) and Fiber Channel over Ethernet (FCoE).

    Switching technology for traditional data centers is generally gigabit. Limited processing capability of servers and low bandwidth I/O interfac restrains the development of high-density high-performance gigabit ports.

    DCE is specifically designed for new generation data centers; it improves legacy Ethernet by enabling high performance, low latency, high cost-effectiveness, zero packet loss, and priority flow control mechanism. With support for L2 multipath Ethernet, IEEE DCE standards are not only necessary for lossless Ethernet in super large-scale data centers, but also for simplifying data centers and integrating internal clouds.

    Existing data centers include IP local network, fiber storage network, and high-performance computing network. However, these networks have different bridging standards. IP local networks adopt Ethernet; fiber storage networks use optical fiber channels; and high-performance computing networks employ HyperLink. Under these conditions, servers require various I/O cards for network management and backup, and are connected to form a complex structure.

    FCoE is used to map and transport optical fiber channel frames over Ethernet. It enables the frames to operate over data center Ethernet without any loss. FCoE enables Optical storage and data Ethernet share the same port, through which both Local Area Network (LAN) and Storage Area Network (SAN) connect to the server. This greatly reduces the number of I/O adapters and cables.

    A unified fabric based on DCE and FCoE allows access to all resources in the LAN, SAN, and high-performance network over lossless 10G Ethernet. In such a fabric, physical network resources are consolidated, and the number of devices, network cards, adapters, switches, and cables are reduced. Power and cooling requirements are also reduced, network architecture is optimized, and maintenance and administration is simplified.

    (2) Unified Virtualization
    Virtualization of computing platform and server enables upper layer applications to schedule CPU, memory, I/O, and application functions on demand and without regard for physical associations or locations. VMWare’s VMotion is the most successful commercial virtualization solution for x86 based servers; Microsoft, Intel, and AMD also provide solutions.

    When hardware resources are virtualized by software, and when a server is virtualized into multiple servers, operation, management, and policy of a data center becomes highly complex. Virtualization brings efficiency benefits, but also creates challenges in complexity of management flows, and software consistence, verification, and security.

    It is becoming clear that, besides application, host, and operating system, network plays an important role in server virtualization. The network can be used in data processing, but also for a storage or computing. It bridges various resources and facilitates resource virtualization. Cloud computing requires ubiquitous data centers, while server virtualization relies on Virtual Machine (VM) migration technology for resource sharing and multiplexing. VM migration requires a unified cross-VM virtualized network environment.

    The network is the basis for guaranteeing QoS and security of rich media cloud services in an enterprise’s internal cloud. Therefore, virtualization should be four dimensional; it should be implemented on the computing layer, storage layer, application layer, and network layer. More importantly, it should be end-to-end and integrated. This is the essential difference between internal rich media clouds and early public stand-alone clouds.

    Integrating computing virtualization, application virtualization, and network virtualization is very important. Without a unified virtualization platform, monitoring and executing various VM-based network and storage policies is difficult. Internal cloud scalability is also constrained, and cross-network virtualization is difficult to implement when multiple physical machines execute the same application.

    VN-link technology has been proposed as a standard. VN-link helps ensure mutual association and communication between network and server with virtualization changes. In this way, the network can distinguish which VM a sent message comes from, and can provide services according to the VM and related policy. An appropriate mechanism for tracking VM migration guarantees service consistency.

    (3) Unified Computing System (UCS)
    Cloud computing implementations are treated as isolated virtual application islands, with isolated platforms and parts. In cloud computing, virtualization increases complexity; and isolated islands challenge multipoint integration and management, increase maintenance costs and risk, and degrade deployment of applications and services.

    Using a unified computing system to integrate and manage computing, network, storage, and virtualization resources is crucial for the implementation of internal enterprise and rich media clouds. Unified systems help reduce ICT infrastructure costs and O&M complexity. They enhance ICT service flexibility, meeting user requirements for future service development.
In a UCS, servers are integrated into the network platform. Intel Nehalem B-series blade servers are required. The patented enhanced memory technology provided by blade servers increases the number of VMs that can be supported by every server. UCS provides unified external interfaces to storage LAN and Network-Attached Storage (NAS) systems. Users can access storage through Ethernet, fiber channels, Ethernet fiber channels, or Internet Small Computer Systems Interface (iSCSI). In this way, UCS provides maximum protection for investment. Only one data center switch is required in UCS, greatly reducing the number of devices, I/O interfaces, and cables, and lowering maintenance cost. Fewer devices mean that an integrated data center is competitively priced, and Total Cost of Ownership (TCO) to clients is greatly reduced.

    In UCS, management functions are integrated into all the system components, and the UCS manager provides holistic solutions as a single entity. The UCS manager also provides a Graphical User Interface (GUI), a Command-Line Interface (CLI), and an Application Programming Interface (API). Management capability is the most important component of UCS, and VM management is at the core of management capability.

    UCS is representative of the transition of data centers from traditional cloud computing to future cloud computing. It has next generation data center networking concepts and technologies such as data center platform and DCE/FCoE unified fabrics. Using the unified switching platform, applications including large-scale high-performance computing and search engine can be implemented and optimized. Unnecessary switches, network cards, cables, and management modules are discarded so that UCS can implement component-integrated cloud platform architecture with unified network, unified virtualization, unified computing, and unified management.

4 Operators and Cloud Computing
    Cloud computing is rewriting the rules in IT, communications, and the Internet. The rapid growth of Internet traffic and the emergence of Internet applications has greatly affected telecom services. Web applications are increasingly visual, community-centered, and personalized, and their operators are concerned with investment return and flexibility of service provision.

    Traditionally, when a telecom operator wished to deliver a new service, an individual network was built. Isolated service islands are in this way formed; resource use is not optimized, and O&M is complicated.

    Cloud computing creates both challenges and opportunities for telecom operators. With abundant network bandwidth and various hardware and software resources in their large-scale data centers, operators usually have the ability to build internal cloud architecture and to develop competitive rich media services. However, in order to successfully transform their services, they should not only play to traditional strengths but also use advanced technologies.

    Existing data or service centers should be integrated into a single new data or service center, and full use should be made of existing next generation networks. A new data or service center and next generation network are IP-based, and further integration creates a powerful and flexible unified service implementation system. This system becomes not only a unified implementation platform for telephone, video, mobile, and data services, but also supports rich media cloud applications and services of the future. In contrast to platforms for traditional services, rich media cloud platforms associate the data center with service transport network in order to provide safer and better quality applications. The introduction of unified rich media cloud platforms will address the important issues of lowering operation costs and speeding up commercialization of services during telecom transformation.

5 The Future of Cloud Computing
    The goal of cloud computing is to develop cross-operator services that rely absolutely on open standards.  Regardless of whether clouds are stand-alone or rich media, they should be able to randomly migrate from one operator to another. Using common open standards, Inter-cloud is designed for flexible scheduling of cloud applications across differ

[Abstract] Cloud computing is a new network computing paradigm based on IP architecture, and its potential lies in new ICT business applications. For the majority of operators and enterprises, the main task associated with cloud computing is next generation data center transformation. This will ensure cloud computing becomes more widespread among enterprises, institutions, organizations, and operators. Cloud computing not only provides traditional IT resource usage and application services, but also supports full resource usage and application services such as IT, communications, video, mobile, and Internet of Things using a converged network infrastructure. Key cloud computing technologies include unified fabric, unified virtualization, and unified computing system. The formation of an open industry alliance and promotion of open technology standards will be critical for the future development of cloud computing.

[Keywords] cloud computing; unified fabrics; unified virtualization; unified computing system