CC Unit I.pdf

CHAPTER Introduction 1 Computing is being transformed into a model consisting of services that are commoditi...

CHAPTER Introduction 1 Computing is being transformed into a model consisting of services that are commoditized and delivered in a manner similar to utilities such as water, electricity, gas, and telephony. In such a model, users access services based on their requirements, regardless of where the services are hosted. Several computing paradigms, such as grid computing, have promised to deliver this utility computing vision. Cloud computing is the most recent emerging paradigm promising to turn the vision of “computing utilities” into a reality. Cloud computing is a technological advancement that focuses on the way we design computing systems, develop applications, and leverage existing services for building software. It is based on the concept of dynamic provisioning, which is applied not only to services but also to compute capability, storage, networking, and information technology (IT) infrastructure in general. Resources are made available through the Internet and offered on a pay-per-use basis from cloud computing vendors. Today, anyone with a credit card can subscribe to cloud services and deploy and configure servers for an application in hours, growing and shrinking the infrastructure serving its application according to the demand, and paying only for the time these resources have been used. This chapter provides a brief overview of the cloud computing phenomenon by presenting its vision, discussing its core features, and tracking the technological developments that have made it possible. The chapter also introduces some key cloud computing technologies as well as some insights into development of cloud computing environments. 1.1 Cloud computing at a glance In 1969, Leonard Kleinrock, one of the chief scientists of the original Advanced Research Projects Agency Network (ARPANET), which seeded the Internet, said: As of now, computer networks are still in their infancy, but as they grow up and become sophisti- cated, we will probably see the spread of ‘computer utilities’ which, like present electric and tele- phone utilities, will service individual homes and offices across the country. This vision of computing utilities based on a service-provisioning model anticipated the massive transformation of the entire computing industry in the 21st century, whereby computing services will be readily available on demand, just as other utility services such as water, electricity, tele- phone, and gas are available in today’s society. Similarly, users (consumers) need to pay providers 3 4 CHAPTER 1 Introduction only when they access the computing services. In addition, consumers no longer need to invest heavily or encounter difficulties in building and maintaining complex IT infrastructure. In such a model, users access services based on their requirements without regard to where the services are hosted. This model has been referred to as utility computing or, recently (since 2007), as cloud computing. The latter term often denotes the infrastructure as a “cloud” from which busi- nesses and users can access applications as services from anywhere in the world and on demand. Hence, cloud computing can be classified as a new paradigm for the dynamic provisioning of com- puting services supported by state-of-the-art data centers employing virtualization technologies for consolidation and effective utilization of resources. Cloud computing allows renting infrastructure, runtime environments, and services on a pay- per-use basis. This principle finds several practical applications and then gives different images of cloud computing to different people. Chief information and technology officers of large enterprises see opportunities for scaling their infrastructure on demand and sizing it according to their business needs. End users leveraging cloud computing services can access their documents and data anytime, anywhere, and from any device connected to the Internet. Many other points of view exist.1 One of the most diffuse views of cloud computing can be summarized as follows: I don’t care where my servers are, who manages them, where my documents are stored, or where my applications are hosted. I just want them always available and access them from any device connected through Internet. And I am willing to pay for this service for as a long as I need it. The concept expressed above has strong similarities to the way we use other services, such as water and electricity. In other words, cloud computing turns IT services into utilities. Such a deliv- ery model is made possible by the effective composition of several technologies, which have reached the appropriate maturity level. Web 2.0 technologies play a central role in making cloud computing an attractive opportunity for building computing systems. They have transformed the Internet into a rich application and service delivery platform, mature enough to serve complex needs. Service orientation allows cloud computing to deliver its capabilities with familiar abstrac- tions, while virtualization confers on cloud computing the necessary degree of customization, con- trol, and flexibility for building production and enterprise systems. Besides being an extremely flexible environment for building new systems and applications, cloud computing also provides an opportunity for integrating additional capacity or new features into existing systems. The use of dynamically provisioned IT resources constitutes a more attractive opportunity than buying additional infrastructure and software, the sizing of which can be difficult to estimate and the needs of which are limited in time. This is one of the most important advan- tages of cloud computing, which has made it a popular phenomenon. With the wide deployment of cloud computing systems, the foundation technologies and systems enabling them are becoming consolidated and standardized. This is a fundamental step in the realization of the long-term vision 1 An interesting perspective on the way cloud computing evokes different things to different people can be found in a series of interviews made by Rob Boothby, vice president and platform evangelist of Joyent, at the Web 2.0 Expo in May 2007. Chief executive officers (CEOs), chief technology officers (CTOs), founders of IT companies, and IT ana- lysts were interviewed, and all of them gave their personal perception of the phenomenon, which at that time was start- ing to spread. The video of the interview can be found on YouTube at the following link: www.youtube.com/watch? v56PNuQHUiV3Q.  1.1 Cloud computing at a glance 5 for cloud computing, which provides an open environment where computing, storage, and other ser- vices are traded as computing utilities. 1.1.1 The vision of cloud computing Cloud computing allows anyone with a credit card to provision virtual hardware, runtime environ- ments, and services. These are used for as long as needed, with no up-front commitments required. The entire stack of a computing system is transformed into a collection of utilities, which can be provisioned and composed together to deploy systems in hours rather than days and with virtually no maintenance costs. This opportunity, initially met with skepticism, has now become a practice across several application domains and business sectors (see Figure 1.1). The demand has fast- tracked technical development and enriched the set of services offered, which have also become more sophisticated and cheaper. Despite its evolution, the use of cloud computing is often limited to a single service at a time or, more commonly, a set of related services offered by the same vendor. Previously, the lack of effective standardization efforts made it difficult to move hosted services from one vendor to another. The long-term vision of cloud computing is that IT services are traded as utilities in an open market, without technological and legal barriers. In this cloud marketplace, cloud service pro- viders and consumers, trading cloud services as utilities, play a central role. Many of the technological elements contributing to this vision already exist. Different stake- holders leverage clouds for a variety of services. The need for ubiquitous storage and compute power on demand is the most common reason to consider cloud computing. A scalable runtime for applications is an attractive option for application and system developers that do not have infra- structure or cannot afford any further expansion of existing infrastructure. The capability for Web- based access to documents and their processing using sophisticated applications is one of the appealing factors for end users. In all these cases, the discovery of such services is mostly done by human intervention: a person (or a team of people) looks over the Internet to identify offerings that meet his or her needs. We imagine that in the near future it will be possible to find the solution that matches our needs by simply entering our request in a global digital market that trades cloud computing services. The existence of such a market will enable the automation of the discovery process and its integration into existing software systems, thus allowing users to transparently leverage cloud resources in their applications and systems. The existence of a global platform for trading cloud services will also help service providers become more visible and therefore potentially increase their revenue. A global cloud market also reduces the barriers between service consumers and providers: it is no lon- ger necessary to belong to only one of these two categories. For example, a cloud provider might become a consumer of a competitor service in order to fulfill its own promises to customers. These are all possibilities that are introduced with the establishment of a global cloud comput- ing marketplace and by defining effective standards for the unified representation of cloud services as well as the interaction among different cloud technologies. A considerable shift toward cloud computing has already been registered, and its rapid adoption facilitates its consolidation. Moreover, by concentrating the core capabilities of cloud computing into large datacenters, it is possible to reduce or remove the need for any technical infrastructure on the service consumer side. This approach provides opportunities for optimizing datacenter facilities and fully utilizing their  I need to grow my I have a lot of infrastructure, but infrastructure that I I do not know for want to rent … how long… I have a surplus of infrastructure that I want to make use of I cannot invest in infrastructure, I just started my business…. I have infrastructure and middleware and I can host applications I want to focus on application logic and not maintenance and scalability issues I have infrastructure and provide application services I want to access and edit my documents and photos from everywhere.. FIGURE 1.1 Cloud computing vision.  1.1 Cloud computing at a glance 7 capabilities to serve multiple users. This consolidation model will reduce the waste of energy and carbon emissions, thus contributing to a greener IT on one end and increasing revenue on the other end. 1.1.2 Defining a cloud Cloud computing has become a popular buzzword; it has been widely used to refer to different technologies, services, and concepts. It is often associated with virtualized infrastructure or hard- ware on demand, utility computing, IT outsourcing, platform and software as a service, and many other things that now are the focus of the IT industry. Figure 1.2 depicts the plethora of different notions included in current definitions of cloud computing. The term cloud has historically been used in the telecommunications industry as an abstraction of the network in system diagrams. It then became the symbol of the most popular computer network: the Internet. This meaning also applies to cloud computing, which refers to an Internet-centric way of No capital investments Clou dbus r ting SaaS Quality of Service Inte S Gree Pay as you go rne Paa n com t putin g Billing g IaaS putin y com evel Utilit Elas cing ice L ticity sour Serv ment IT out Agre e l tua Vir nters lity IT outsourcing e labi Da ta c Sca Privac y & Tru g st onin visi d Pro eman Security on d tion aliza Vir tu FIGURE 1.2 Cloud computing technologies, concepts, and ideas. 8 CHAPTER 1 Introduction computing. The Internet plays a fundamental role in cloud computing, since it represents either the medium or the platform through which many cloud computing services are delivered and made accessible. This aspect is also reflected in the definition given by Armbrust et al. : Cloud computing refers to both the applications delivered as services over the Internet and the hardware and system software in the datacenters that provide those services. This definition describes cloud computing as a phenomenon touching on the entire stack: from the underlying hardware to the high-level software services and applications. It introduces the con- cept of everything as a service, mostly referred as XaaS,2 where the different components of a sys- tem—IT infrastructure, development platforms, databases, and so on—can be delivered, measured, and consequently priced as a service. This new approach significantly influences not only the way that we build software but also the way we deploy it, make it accessible, and design our IT infra- structure, and even the way companies allocate the costs for IT needs. The approach fostered by cloud computing is global: it covers both the needs of a single user hosting documents in the cloud and the ones of a CIO deciding to deploy part of or the entire corporate IT infrastructure in the pub- lic cloud. This notion of multiple parties using a shared cloud computing environment is highlighted in a definition proposed by the U.S. National Institute of Standards and Technology (NIST): Cloud computing is a model for enabling ubiquitous, convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, servers, storage, applications, and services) that can be rapidly provisioned and released with minimal management effort or service provider interaction. Another important aspect of cloud computing is its utility-oriented approach. More than any other trend in distributed computing, cloud computing focuses on delivering services with a given pricing model, in most cases a “pay-per-use” strategy. It makes it possible to access online storage, rent virtual hardware, or use development platforms and pay only for their effective usage, with no or minimal up-front costs. All these operations can be performed and billed simply by entering the credit card details and accessing the exposed services through a Web browser. This helps us pro- vide a different and more practical characterization of cloud computing. According to Reese , we can define three criteria to discriminate whether a service is delivered in the cloud computing style: The service is accessible via a Web browser (nonproprietary) or a Web services application programming interface (API). Zero capital expenditure is necessary to get started. You pay only for what you use as you use it. Even though many cloud computing services are freely available for single users, enterprise- class services are delivered according a specific pricing scheme. In this case users subscribe to the service and establish with the service provider a service-level agreement (SLA) defining the 2 XaaS is an acronym standing for X-as-a-Service, where the X letter can be replaced by one of a number of things: S for software, P for platform, I for infrastructure, H for hardware, D for database, and so on.  1.1 Cloud computing at a glance 9 quality-of-service parameters under which the service is delivered. The utility-oriented nature of cloud computing is clearly expressed by Buyya et al. : A cloud is a type of parallel and distributed system consisting of a collection of interconnected and virtualized computers that are dynamically provisioned and presented as one or more unified computing resources based on service-level agreements established through negotiation between the service provider and consumers. 1.1.3 A closer look Cloud computing is helping enterprises, governments, public and private institutions, and research organizations shape more effective and demand-driven computing systems. Access to, as well as integration of, cloud computing resources and systems is now as easy as performing a credit card transaction over the Internet. Practical examples of such systems exist across all market segments: Large enterprises can offload some of their activities to cloud-based systems. Recently, the New York Times has converted its digital library of past editions into a Web-friendly format. This required a considerable amount of computing power for a short period of time. By renting Amazon EC2 and S3 Cloud resources, the Times performed this task in 36 hours and relinquished these resources, with no additional costs. Small enterprises and start-ups can afford to translate their ideas into business results more quickly, without excessive up-front costs. Animoto is a company that creates videos out of images, music, and video fragments submitted by users. The process involves a considerable amount of storage and backend processing required for producing the video, which is finally made available to the user. Animoto does not own a single server and bases its computing infrastructure entirely on Amazon Web Services, which are sized on demand according to the overall workload to be processed. Such workload can vary a lot and require instant scalability.3 Up-front investment is clearly not an effective solution for many companies, and cloud computing systems become an appropriate alternative. System developers can concentrate on the business logic rather than dealing with the complexity of infrastructure management and scalability. Little Fluffy Toys is a company in London that has developed a widget providing users with information about nearby bicycle rental services. The company has managed to back the widget’s computing needs on Google AppEngine and be on the market in only one week. End users can have their documents accessible from everywhere and any device. Apple iCloud is a service that allows users to have their documents stored in the Cloud and access them from any device users connect to it. This makes it possible to take a picture while traveling with a smartphone, go back home and edit the same picture on your laptop, and have it show as updated on your tablet computer. This process is completely transparent to the user, who does not have to set up cables and connect these devices with each other. How is all of this made possible? The same concept of IT services on demand—whether com- puting power, storage, or runtime environments for applications—on a pay-as-you-go basis 3 It has been reported that Animoto, in one single week, scaled from 70 to 8,500 servers because of user demand. 10 CHAPTER 1 Introduction accommodates these four different scenarios. Cloud computing does not only contribute with the opportunity of easily accessing IT services on demand, it also introduces a new way of thinking about IT services and resources: as utilities. A bird’s-eye view of a cloud computing environment is shown in Figure 1.3. The three major models for deploying and accessing cloud computing environments are public clouds, private/enterprise clouds, and hybrid clouds (see Figure 1.4). Public clouds are the most common deployment models in which necessary IT infrastructure (e.g., virtualized datacenters) is established by a third-party service provider that makes it available to any consumer on a subscrip- tion basis. Such clouds are appealing to users because they allow users to quickly leverage com- pute, storage, and application services. In this environment, users’ data and applications are deployed on cloud datacenters on the vendor’s premises. Large organizations that own massive computing infrastructures can still benefit from cloud computing by replicating the cloud IT service delivery model in-house. This idea has given birth to the concept of private clouds as opposed to public clouds. In 2010, for example, the U.S. federal government, one of the world’s largest consumers of IT spending (around $76 billion on more than Subscription - Oriented Cloud Services: X{compute, apps, data,..} Manjrasoft as a Service (..aaS) Public Clouds Applications Development and Runtime Platform Compute Cloud Hy Storage br Manager id Cl ou d Clients Private Cloud Other Govt. Cloud Services Cloud Services FIGURE 1.3 A bird’s-eye view of cloud computing.  1.1 Cloud computing at a glance 11 Cloud Deployment Models Public/Internet Private/Enterprise Hybrid/Inter Clouds Clouds Clouds *Third-party, *A public cloud model * Mixed use of multitenant cloud within a private and public infrastructure company’s own clouds; leasing public and services datacenter/infrastructure cloud services for internal when private cloud *Available on a and/or partners’ use capacity is insufficient subscription basis to all FIGURE 1.4 Major deployment models for cloud computing. 10,000 systems) started a cloud computing initiative aimed at providing government agencies with a more efficient use of their computing facilities. The use of cloud-based in-house solutions is also driven by the need to keep confidential information within an organization’s premises. Institutions such as governments and banks that have high security, privacy, and regulatory concerns prefer to build and use their own private or enterprise clouds. Whenever private cloud resources are unable to meet users’ quality-of-service requirements, hybrid computing systems, partially composed of public cloud resources and privately owned infra- structures, are created to serve the organization’s needs. These are often referred as hybrid clouds, which are becoming a common way for many stakeholders to start exploring the possibilities offered by cloud computing. 1.1.4 The cloud computing reference model A fundamental characteristic of cloud computing is the capability to deliver, on demand, a variety of IT services that are quite diverse from each other. This variety creates different perceptions of what cloud computing is among users. Despite this lack of uniformity, it is possible to classify cloud computing services offerings into three major categories: Infrastructure-as-a-Service (IaaS), Platform-as-a-Service (PaaS), and Software-as-a-Service (SaaS). These categories are related to each other as described in Figure 1.5, which provides an organic view of cloud computing. We refer to this diagram as the Cloud Computing Reference Model, and we will use it throughout the 12 CHAPTER 1 Introduction Web 2.0 Software as a Service Interfaces End-user applications Scientific applications Office automation, photo editing, CRM, and social networking Examples : Google Documents, Facebook, Flickr, Salesforce Platform as a Service Runtime environment for applications Development and data processing platforms Examples : Windows Azure, Hadoop, Google AppEngine, Aneka Infrastructure as a Service Virtualized servers Storage and networking Examples : Amazon EC2, S3, Rightscale, vCloud FIGURE 1.5 The Cloud Computing Reference Model. book to explain the technologies and introduce the relevant research on this phenomenon. The model organizes the wide range of cloud computing services into a layered view that walks the computing stack from bottom to top. At the base of the stack, Infrastructure-as-a-Service solutions deliver infrastructure on demand in the form of virtual hardware, storage, and networking. Virtual hardware is utilized to provide compute on demand in the form of virtual machine instances. These are created at users’ request on the provider’s infrastructure, and users are given tools and interfaces to configure the software stack installed in the virtual machine. The pricing model is usually defined in terms of dollars per hour, where the hourly cost is influenced by the characteristics of the virtual hardware. Virtual storage is delivered in the form of raw disk space or object store. The former complements a virtual hardware offering that requires persistent storage. The latter is a more high-level abstraction for storing enti- ties rather than files. Virtual networking identifies the collection of services that manage the net- working among virtual instances and their connectivity to the Internet or private networks. Platform-as-a-Service solutions are the next step in the stack. They deliver scalable and elastic runtime environments on demand and host the execution of applications. These services are backed by a core middleware platform that is responsible for creating the abstract environment where applications are deployed and executed. It is the responsibility of the service provider to provide scalability and to manage fault tolerance, while users are requested to focus on the logic of the application developed by leveraging the provider’s APIs and libraries. This approach increases the level of abstraction at which cloud computing is leveraged but also constrains the user in a more controlled environment. At the top of the stack, Software-as-a-Service solutions provide applications and services on demand. Most of the common functionalities of desktop applications—such as office  1.1 Cloud computing at a glance 13 automation, document management, photo editing, and customer relationship management (CRM) software—are replicated on the provider’s infrastructure and made more scalable and accessible through a browser on demand. These applications are shared across multiple users whose interac- tion is isolated from the other users. The SaaS layer is also the area of social networking Websites, which leverage cloud-based infrastructures to sustain the load generated by their popularity. Each layer provides a different service to users. IaaS solutions are sought by users who want to leverage cloud computing from building dynamically scalable computing systems requiring a spe- cific software stack. IaaS services are therefore used to develop scalable Websites or for back- ground processing. PaaS solutions provide scalable programming platforms for developing applications and are more appropriate when new systems have to be developed. SaaS solutions tar- get mostly end users who want to benefit from the elastic scalability of the cloud without doing any software development, installation, configuration, and maintenance. This solution is appropriate when there are existing SaaS services that fit users needs (such as email, document management, CRM, etc.) and a minimum level of customization is needed. 1.1.5 Characteristics and benefits Cloud computing has some interesting characteristics that bring benefits to both cloud service con- sumers (CSCs) and cloud service providers (CSPs). These characteristics are: No up-front commitments On-demand access Nice pricing Simplified application acceleration and scalability Efficient resource allocation Energy efficiency Seamless creation and use of third-party services The most evident benefit from the use of cloud computing systems and technologies is the increased economical return due to the reduced maintenance costs and operational costs related to IT software and infrastructure. This is mainly because IT assets, namely software and infrastructure, are turned into utility costs, which are paid for as long as they are used, not paid for up front. Capital costs are costs associated with assets that need to be paid in advance to start a business activity. Before cloud computing, IT infrastructure and software generated capital costs, since they were paid up front so that business start-ups could afford a computing infrastructure, enabling the business activities of the organization. The revenue of the business is then utilized to compensate over time for these costs. Organizations always minimize capital costs, since they are often associ- ated with depreciable values. This is the case of hardware: a server bought today for $1,000 will have a market value less than its original price when it is eventually replaced by new hardware. To make profit, organizations have to compensate for this depreciation created by time, thus reducing the net gain obtained from revenue. Minimizing capital costs, then, is fundamental. Cloud comput- ing transforms IT infrastructure and software into utilities, thus significantly contributing to increas- ing a company’s net gain. Moreover, cloud computing also provides an opportunity for small organizations and start-ups: these do not need large investments to start their business, but they can comfortably grow with it. Finally, maintenance costs are significantly reduced: by renting the 14 CHAPTER 1 Introduction infrastructure and the application services, organizations are no longer responsible for their mainte- nance. This task is the responsibility of the cloud service provider, who, thanks to economies of scale, can bear the maintenance costs. Increased agility in defining and structuring software systems is another significant benefit of cloud computing. Since organizations rent IT services, they can more dynamically and flexibly com- pose their software systems, without being constrained by capital costs for IT assets. There is a reduced need for capacity planning, since cloud computing allows organizations to react to unplanned surges in demand quite rapidly. For example, organizations can add more servers to pro- cess workload spikes and dismiss them when they are no longer needed. Ease of scalability is another advantage. By leveraging the potentially huge capacity of cloud computing, organizations can extend their IT capability more easily. Scalability can be leveraged across the entire computing stack. Infrastructure providers offer simple methods to provision customized hardware and integrate it into existing systems. Platform-as-a-Service providers offer runtime environment and programming mod- els that are designed to scale applications. Software-as-a-Service offerings can be elastically sized on demand without requiring users to provision hardware or to program application for scalability. End users can benefit from cloud computing by having their data and the capability of operating on it always available, from anywhere, at any time, and through multiple devices. Information and services stored in the cloud are exposed to users by Web-based interfaces that make them accessi- ble from portable devices as well as desktops at home. Since the processing capabilities (that is, office automation features, photo editing, information management, and so on) also reside in the cloud, end users can perform the same tasks that previously were carried out through considerable software investments. The cost for such opportunities is generally very limited, since the cloud ser- vice provider shares its costs across all the tenants that he is servicing. Multitenancy allows for bet- ter utilization of the shared infrastructure that is kept operational and fully active. The concentration of IT infrastructure and services into large datacenters also provides opportunity for considerable optimization in terms of resource allocation and energy efficiency, which eventually can lead to a less impacting approach on the environment. Finally, service orientation and on-demand access create new opportunities for composing sys- tems and applications with a flexibility not possible before cloud computing. New service offerings can be created by aggregating together existing services and concentrating on added value. Since it is possible to provision on demand any component of the computing stack, it is easier to turn ideas into products with limited costs and by concentrating technical efforts on what matters: the added value. 1.1.6 Challenges ahead As any new technology develops and becomes popular, new issues have to be faced. Cloud com- puting is not an exception. New, interesting problems and challenges are regularly being posed to the cloud community, including IT practitioners, managers, governments, and regulators. Besides the practical aspects, which are related to configuration, networking, and sizing of cloud computing systems, a new set of challenges concerning the dynamic provisioning of cloud comput- ing services and resources arises. For example, in the Infrastructure-as-a-Service domain, how many resources need to be provisioned, and for how long should they be used, in order to maxi- mize the benefit? Technical challenges also arise for cloud service providers for the management of large computing infrastructures and the use of virtualization technologies on top of them. In  1.2 Historical developments 15 addition, issues and challenges concerning the integration of real and virtual infrastructure need to be taken into account from different perspectives, such as security and legislation. Security in terms of confidentiality, secrecy, and protection of data in a cloud environment is another important challenge. Organizations do not own the infrastructure they use to process data and store information. This condition poses challenges for confidential data, which organizations cannot afford to reveal. Therefore, assurance on the confidentiality of data and compliance to secu- rity standards, which give a minimum guarantee on the treatment of information on cloud comput- ing systems, are sought. The problem is not as evident as it seems: even though cryptography can help secure the transit of data from the private premises to the cloud infrastructure, in order to be processed the information needs to be decrypted in memory. This is the weak point of the chain: since virtualization allows capturing almost transparently the memory pages of an instance, these data could easily be obtained by a malicious provider. Legal issues may also arise. These are specifically tied to the ubiquitous nature of cloud com- puting, which spreads computing infrastructure across diverse geographical locations. Different leg- islation about privacy in different countries may potentially create disputes as to the rights that third parties (including government agencies) have to your data. U.S. legislation is known to give extreme powers to government agencies to acquire confidential data when there is the suspicion of operations leading to a threat to national security. European countries are more restrictive and pro- tect the right of privacy. An interesting scenario comes up when a U.S. organization uses cloud ser- vices that store their data in Europe. In this case, should this organization be suspected by the government, it would become difficult or even impossible for the U.S. government to take control of the data stored in a cloud datacenter located in Europe. 1.2 Historical developments The idea of renting computing services by leveraging large distributed computing facilities has been around for long time. It dates back to the days of the mainframes in the early 1950s. From there on, technology has evolved and been refined. This process has created a series of favorable conditions for the realization of cloud computing. Figure 1.6 provides an overview of the evolution of the distributed computing technologies that have influenced cloud computing. In tracking the historical evolution, we briefly review five core technologies that played an important role in the realization of cloud computing. These technolo- gies are distributed systems, virtualization, Web 2.0, service orientation, and utility computing. 1.2.1 Distributed systems Clouds are essentially large distributed computing facilities that make available their services to third parties on demand. As a reference, we consider the characterization of a distributed system proposed by Tanenbaum et al. : A distributed system is a collection of independent computers that appears to its users as a single coherent system.  CHAPTER Virtualization 3 Virtualization technology is one of the fundamental components of cloud computing, especially in regard to infrastructure-based services. Virtualization allows the creation of a secure, customizable, and isolated execution environment for running applications, even if they are untrusted, without affecting other users’ applications. The basis of this technology is the ability of a computer pro- gram—or a combination of software and hardware—to emulate an executing environment separate from the one that hosts such programs. For example, we can run Windows OS on top of a virtual machine, which itself is running on Linux OS. Virtualization provides a great opportunity to build elastically scalable systems that can provision additional capability with minimum costs. Therefore, virtualization is widely used to deliver customizable computing environments on demand. This chapter discusses the fundamental concepts of virtualization, its evolution, and various models and technologies used in cloud computing environments. 3.1 Introduction Virtualization is a large umbrella of technologies and concepts that are meant to provide an abstract environment—whether virtual hardware or an operating system—to run applications. The term virtualization is often synonymous with hardware virtualization, which plays a fundamental role in efficiently delivering Infrastructure-as-a-Service (IaaS) solutions for cloud computing. In fact, virtualization technologies have a long trail in the history of computer science and have been available in many flavors by providing virtual environments at the operating system level, the programming lan- guage level, and the application level. Moreover, virtualization technologies provide a virtual environ- ment for not only executing applications but also for storage, memory, and networking. Since its inception, virtualization has been sporadically explored and adopted, but in the last few years there has been a consistent and growing trend to leverage this technology. Virtualization technologies have gained renewed interested recently due to the confluence of several phenomena: Increased performance and computing capacity. Nowadays, the average end-user desktop PC is powerful enough to meet almost all the needs of everyday computing, with extra capacity that is rarely used. Almost all these PCs have resources enough to host a virtual machine manager and execute a virtual machine with by far acceptable performance. The same consideration applies to the high-end side of the PC market, where supercomputers can provide immense compute power that can accommodate the execution of hundreds or thousands of virtual machines. Underutilized hardware and software resources. Hardware and software underutilization is occurring due to (1) increased performance and computing capacity, and (2) the effect of 71 72 CHAPTER 3 Virtualization limited or sporadic use of resources. Computers today are so powerful that in most cases only a fraction of their capacity is used by an application or the system. Moreover, if we consider the IT infrastructure of an enterprise, many computers are only partially utilized whereas they could be used without interruption on a 24/7/365 basis. For example, desktop PCs mostly devoted to office automation tasks and used by administrative staff are only used during work hours, remaining completely unused overnight. Using these resources for other purposes after hours could improve the efficiency of the IT infrastructure. To transparently provide such a service, it would be necessary to deploy a completely separate environment, which can be achieved through virtualization. Lack of space. The continuous need for additional capacity, whether storage or compute power, makes data centers grow quickly. Companies such as Google and Microsoft expand their infrastructures by building data centers as large as football fields that are able to host thousands of nodes. Although this is viable for IT giants, in most cases enterprises cannot afford to build another data center to accommodate additional resource capacity. This condition, along with hardware underutilization, has led to the diffusion of a technique called server consolidation,1 for which virtualization technologies are fundamental. Greening initiatives. Recently, companies are increasingly looking for ways to reduce the amount of energy they consume and to reduce their carbon footprint. Data centers are one of the major power consumers; they contribute consistently to the impact that a company has on the environment. Maintaining a data center operation not only involves keeping servers on, but a great deal of energy is also consumed in keeping them cool. Infrastructures for cooling have a significant impact on the carbon footprint of a data center. Hence, reducing the number of servers through server consolidation will definitely reduce the impact of cooling and power consumption of a data center. Virtualization technologies can provide an efficient way of consolidating servers. Rise of administrative costs. Power consumption and cooling costs have now become higher than the cost of IT equipment. Moreover, the increased demand for additional capacity, which translates into more servers in a data center, is also responsible for a significant increment in administrative costs. Computers—in particular, servers—do not operate all on their own, but they require care and feeding from system administrators. Common system administration tasks include hardware monitoring, defective hardware replacement, server setup and updates, server resources monitoring, and backups. These are labor-intensive operations, and the higher the number of servers that have to be managed, the higher the administrative costs. Virtualization can help reduce the number of required servers for a given workload, thus reducing the cost of the administrative personnel. These can be considered the major causes for the diffusion of hardware virtualization solutions as well as the other kinds of virtualization. The first step toward consistent adoption of virtualiza- tion technologies was made with the wide spread of virtual machine-based programming languages: In 1995 Sun released Java, which soon became popular among developers. The ability to integrate small Java applications, called applets, made Java a very successful platform, and with the 1 Server consolidation is a technique for aggregating multiple services and applications originally deployed on different servers on one physical server. Server consolidation allows us to reduce the power consumption of a data center and resolve hardware underutilization.  3.2 Characteristics of virtualized environments 73 beginning of the new millennium Java played a significant role in the application server market segment, thus demonstrating that the existing technology was ready to support the execution of managed code for enterprise-class applications. In 2002 Microsoft released the first version of.NET Framework, which was Microsoft’s alternative to the Java technology. Based on the same principles as Java, able to support multiple programming languages, and featuring complete integra- tion with other Microsoft technologies,.NET Framework soon became the principal development platform for the Microsoft world and quickly became popular among developers. In 2006, two of the three “official languages” used for development at Google, Java and Python, were based on the virtual machine model. This trend of shifting toward virtualization from a programming language perspective demonstrated an important fact: The technology was ready to support virtualized solu- tions without a significant performance overhead. This paved the way to another and more radical form of virtualization that now has become a fundamental requisite for any data center management infrastructure. 3.2 Characteristics of virtualized environments Virtualization is a broad concept that refers to the creation of a virtual version of something, whether hardware, a software environment, storage, or a network. In a virtualized environment there are three major components: guest, host, and virtualization layer. The guest represents the system component that interacts with the virtualization layer rather than with the host, as would normally happen. The host represents the original environment where the guest is supposed to be managed. The virtualization layer is responsible for recreating the same or a different environment where the guest will operate (see Figure 3.1). Such a general abstraction finds different applications and then implementations of the virtuali- zation technology. The most intuitive and popular is represented by hardware virtualization, which also constitutes the original realization of the virtualization concept.2 In the case of hardware vir- tualization, the guest is represented by a system image comprising an operating system and installed applications. These are installed on top of virtual hardware that is controlled and managed by the virtualization layer, also called the virtual machine manager. The host is instead represented by the physical hardware, and in some cases the operating system, that defines the environment where the virtual machine manager is running. In the case of virtual storage, the guest might be cli- ent applications or users that interact with the virtual storage management software deployed on top of the real storage system. The case of virtual networking is also similar: The guest— applications and users—interacts with a virtual network, such as a virtual private network (VPN), which is managed by specific software (VPN client) using the physical network available on the node. VPNs are useful for creating the illusion of being within a different physical network and thus accessing the resources in it, which would otherwise not be available. 2 Virtualization is a technology that was initially developed during the mainframe era. The IBM CP/CMS mainframes were the first systems to introduce the concept of hardware virtualization and hypervisors. These systems, able to run multiple operating systems at the same time, provided a backward-compatible environment that allowed customers to run previous versions of their applications. 74 CHAPTER 3 Virtualization Virtual Image Applications Applications Guest Virtual Hardware Virtual Storage Virtual Networking Virtualization Layer Software Emulation Host Physical Hardware Physical Storage Physical Networking FIGURE 3.1 The virtualization reference model. The main common characteristic of all these different implementations is the fact that the vir- tual environment is created by means of a software program. The ability to use software to emulate such a wide variety of environments creates a lot of opportunities, previously less attractive because of excessive overhead introduced by the virtualization layer. The technologies of today allow profitable use of virtualization and make it possible to fully exploit the advantages that come with it. Such advantages have always been characteristics of virtualized solutions. 3.2.1 Increased security The ability to control the execution of a guest in a completely transparent manner opens new pos- sibilities for delivering a secure, controlled execution environment. The virtual machine repre- sents an emulated environment in which the guest is executed. All the operations of the guest are generally performed against the virtual machine, which then translates and applies them to the host. This level of indirection allows the virtual machine manager to control and filter the activity of the guest, thus preventing some harmful operations from being performed. Resources exposed by the host can then be hidden or simply protected from the guest. Moreover, sensitive  3.2 Characteristics of virtualized environments 75 Virtual Resources Sharing Aggregation Emulation Isolation Virtualization Physical Resources FIGURE 3.2 Functions enabled by managed execution. information that is contained in the host can be naturally hidden without the need to install com- plex security policies. Increased security is a requirement when dealing with untrusted code. For example, applets downloaded from the Internet run in a sandboxed3 version of the Java Virtual Machine (JVM), which provides them with limited access to the hosting operating system resources. Both the JVM and the.NET runtime provide extensive security policies for customiz- ing the execution environment of applications. Hardware virtualization solutions such as VMware Desktop, VirtualBox, and Parallels provide the ability to create a virtual computer with custom- ized virtual hardware on top of which a new operating system can be installed. By default, the file system exposed by the virtual computer is completely separated from the one of the host machine. This becomes the perfect environment for running applications without affecting other users in the environment. 3.2.2 Managed execution Virtualization of the execution environment not only allows increased security, but a wider range of features also can be implemented. In particular, sharing, aggregation, emulation, and isolation are the most relevant features (see Figure 3.2). Sharing. Virtualization allows the creation of a separate computing environments within the same host. In this way it is possible to fully exploit the capabilities of a powerful guest, which would otherwise be underutilized. As we will see in later chapters, sharing is a particularly important feature in virtualized data centers, where this basic feature is used to reduce the number of active servers and limit power consumption. 3 The term sandbox identifies an isolated execution environment where instructions can be filtered and blocked before being translated and executed in the real execution environment. The expression sandboxed version of the Java Virtual Machine (JVM) refers to a particular configuration of the JVM where, by means of security policy, instructions that are considered potential harmful can be blocked. 76 CHAPTER 3 Virtualization Aggregation. Not only is it possible to share physical resource among several guests, but virtualization also allows aggregation, which is the opposite process. A group of separate hosts can be tied together and represented to guests as a single virtual host. This function is naturally implemented in middleware for distributed computing, with a classical example represented by cluster management software, which harnesses the physical resources of a homogeneous group of machines and represents them as a single resource. Emulation. Guest programs are executed within an environment that is controlled by the virtualization layer, which ultimately is a program. This allows for controlling and tuning the environment that is exposed to guests. For instance, a completely different environment with respect to the host can be emulated, thus allowing the execution of guest programs requiring specific characteristics that are not present in the physical host. This feature becomes very useful for testing purposes, where a specific guest has to be validated against different platforms or architectures and the wide range of options is not easily accessible during development. Again, hardware virtualization solutions are able to provide virtual hardware and emulate a particular kind of device such as Small Computer System Interface (SCSI) devices for file I/O, without the hosting machine having such hardware installed. Old and legacy software that does not meet the requirements of current systems can be run on emulated hardware without any need to change the code. This is possible either by emulating the required hardware architecture or within a specific operating system sandbox, such as the MS-DOS mode in Windows 95/98. Another example of emulation is an arcade-game emulator that allows us to play arcade games on a normal personal computer. Isolation. Virtualization allows providing guests—whether they are operating systems, applications, or other entities—with a completely separate environment, in which they are executed. The guest program performs its activity by interacting with an abstraction layer, which provides access to the underlying resources. Isolation brings several benefits; for example, it allows multiple guests to run on the same host without interfering with each other. Second, it provides a separation between the host and the guest. The virtual machine can filter the activity of the guest and prevent harmful operations against the host. Besides these characteristics, another important capability enabled by virtualization is perfor- mance tuning. This feature is a reality at present, given the considerable advances in hardware and software supporting virtualization. It becomes easier to control the performance of the guest by finely tuning the properties of the resources exposed through the virtual environment. This capability provides a means to effectively implement a quality-of-service (QoS) infrastructure that more easily fulfills the service-level agreement (SLA) established for the guest. For instance, software-implementing hardware virtualization solutions can expose to a guest operating system only a fraction of the memory of the host machine or set the maximum frequency of the proces- sor of the virtual machine. Another advantage of managed execution is that sometimes it allows easy capturing of the state of the guest program, persisting it, and resuming its execution. This, for example, allows virtual machine managers such as Xen Hypervisor to stop the execution of a guest operating system, move its virtual image into another machine, and resume its execution in a completely transparent manner. This technique is called virtual machine migration and constitu- tes an important feature in virtualized data centers for optimizing their efficiency in serving application demands.  3.3 Taxonomy of virtualization techniques 77 3.2.3 Portability The concept of portability applies in different ways according to the specific type of virtualization con- sidered. In the case of a hardware virtualization solution, the guest is packaged into a virtual image that, in most cases, can be safely moved and executed on top of different virtual machines. Except for the file size, this happens with the same simplicity with which we can display a picture image in differ- ent computers. Virtual images are generally proprietary formats that require a specific virtual machine manager to be executed. In the case of programming-level virtualization, as implemented by the JVM or the.NET runtime, the binary code representing application components (jars or assemblies) can be run without any recompilation on any implementation of the corresponding virtual machine. This makes the application development cycle more flexible and application deployment very straightfor- ward: One version of the application, in most cases, is able to run on different platforms with no changes. Finally, portability allows having your own system always with you and ready to use as long as the required virtual machine manager is available. This requirement is, in general, less stringent than having all the applications and services you need available to you anywhere you go. 3.3 Taxonomy of virtualization techniques Virtualization covers a wide range of emulation techniques that are applied to different areas of computing. A classification of these techniques helps us better understand their characteristics and use (see Figure 3.3). The first classification discriminates against the service or entity that is being emulated. Virtualization is mainly used to emulate execution environments, storage, and networks. Among these categories, execution virtualization constitutes the oldest, most popular, and most developed area. Therefore, it deserves major investigation and a further categorization. In particular we can divide these execution virtualization techniques into two major categories by considering the type of host they require. Process-level techniques are implemented on top of an existing operating sys- tem, which has full control of the hardware. System-level techniques are implemented directly on hardware and do not require—or require a minimum of support from—an existing operating system. Within these two categories we can list various techniques that offer the guest a different type of virtual computation environment: bare hardware, operating system resources, low-level programming language, and application libraries. 3.3.1 Execution virtualization Execution virtualization includes all techniques that aim to emulate an execution environment that is separate from the one hosting the virtualization layer. All these techniques concentrate their inter- est on providing support for the execution of programs, whether these are the operating system, a binary specification of a program compiled against an abstract machine model, or an application. Therefore, execution virtualization can be implemented directly on top of the hardware by the operating system, an application, or libraries dynamically or statically linked to an application image. 78 CHAPTER 3 Virtualization How it is done? Technique Virtualization Model Emulation Application Execution Programming Environment Process Level High-Level VM Language Storage Operating Multiprogramming System Virtualization Network Hardware-Assisted Virtualization Full Virtualization System Level Hardware …. Paravirtualization Partial Virtualization FIGURE 3.3 A taxonomy of virtualization techniques. 3.3.1.1 Machine reference model Virtualizing an execution environment at different levels of the computing stack requires a refer- ence model that defines the interfaces between the levels of abstractions, which hide implementa- tion details. From this perspective, virtualization techniques actually replace one of the layers and intercept the calls that are directed toward it. Therefore, a clear separation between layers simplifies their implementation, which only requires the emulation of the interfaces and a proper interaction with the underlying layer. Modern computing systems can be expressed in terms of the reference model described in Figure 3.4. At the bottom layer, the model for the hardware is expressed in terms of the Instruction Set Architecture (ISA), which defines the instruction set for the processor, registers, memory, and interrupt management. ISA is the interface between hardware and software, and it is important to the operating system (OS) developer (System ISA) and developers of applications that directly manage the underlying hardware (User ISA). The application binary interface (ABI) separates the operating system layer from the applica- tions and libraries, which are managed by the OS. ABI covers details such as low-level data types, align- ment, and call conventions and defines a format for executable programs. System calls are defined at this level. This interface allows portability of applications and libraries across operating systems that  3.3 Taxonomy of virtualization techniques 79 Applications Applications API calls API Libraries Libraries ABI System calls User User ISA Operative System Operative System ISA ISA ISA Hardware Hardware FIGURE 3.4 A machine reference model. implement the same ABI. The highest level of abstraction is represented by the application programming interface (API), which interfaces applications to libraries and/or the underlying operating system. For any operation to be performed in the application level API, ABI and ISA are responsible for making it happen. The high-level abstraction is converted into machine-level instructions to per- form the actual operations supported by the processor. The machine-level resources, such as proces- sor registers and main memory capacities, are used to perform the operation at the hardware level of the central processing unit (CPU). This layered approach simplifies the development and imple- mentation of computing systems and simplifies the implementation of multitasking and the coexis- tence of multiple executing environments. In fact, such a model not only requires limited knowledge of the entire computing stack, but it also provides ways to implement a minimal security model for managing and accessing shared resources. For this purpose, the instruction set exposed by the hardware has been divided into different security classes that define who can operate with them. The first distinction can be made between privileged and nonprivileged instructions. Nonprivileged instructions are those instructions that can be used without interfering with other tasks because they do not access shared resources. This cate- gory contains, for example, all the floating, fixed-point, and arithmetic instructions. Privileged instructions are those that are executed under specific restrictions and are mostly used for sensitive operations, which expose (behavior-sensitive) or modify (control-sensitive) the privileged state. For instance, behavior-sensitive instructions are those that operate on the I/O, whereas control-sensitive instructions alter the state of the CPU registers. Some types of architecture feature more than one class of privileged instructions and implement a finer control of how these instructions can be accessed. For instance, a possible implementation features a hierarchy of privileges (see Figure 3.5) in the form of ring-based security: Ring 0, Ring 1, Ring 2, and Ring 3; Ring 0 is in the most privi- leged level and Ring 3 in the least privileged level. Ring 0 is used by the kernel of the OS, rings 1 and 2 are used by the OS-level services, and Ring 3 is used by the user. Recent systems support only two levels, with Ring 0 for supervisor mode and Ring 3 for user mode. 80 CHAPTER 3 Virtualization Least Privileged Mode (User Mode) Ring 3 Privileged Modes Ring 2 Ring 1 Ring 0 Most Privileged Mode (Supervisor Mode) FIGURE 3.5 Security rings and privilege modes. All the current systems support at least two different execution modes: supervisor mode and user mode. The first mode denotes an execution mode in which all the instructions (privileged and nonprivi- leged) can be executed without any restriction. This mode, also called master mode or kernel mode, is generally used by the operating system (or the hypervisor) to perform sensitive operations on hardware- level resources. In user mode, there are restrictions to control the machine-level resources. If code run- ning in user mode invokes the privileged instructions, hardware interrupts occur and trap the potentially harmful execution of the instruction. Despite this, there might be some instructions that can be invoked as privileged instructions under some conditions and as nonprivileged instructions under other conditions. The distinction between user and supervisor mode allows us to understand the role of the hypervisor and why it is called that. Conceptually, the hypervisor runs above the supervisor mode, and from here the prefix hyper- is used. In reality, hypervisors are run in supervisor mode, and the division between privileged and nonprivileged instructions has posed challenges in designing virtual machine managers. It is expected that all the sensitive instructions will be executed in privileged mode, which requires supervisor mode in order to avoid traps. Without this assumption it is impossible to fully emulate and manage the status of the CPU for guest operating systems. Unfortunately, this is not true for the original ISA, which allows 17 sensitive instructions to be called in user mode. This prevents multiple operating systems managed by a single hypervisor to be isolated from each other, since they are able to access the privileged state of the processor and change it.4 More recent implementations of ISA (Intel VT and AMD Pacifica) have solved this problem by redesigning such instructions as privileged ones. By keeping in mind this reference model, it is possible to explore and better understand the var- ious techniques utilized to virtualize execution environments and their relationships to the other components of the system. 4 It is expected that in a hypervisor-managed environment, all the guest operating system code will be run in user mode in order to prevent it from directly accessing the status of the CPU. If there are sensitive instructions that can be called in user mode (that is, implemented as nonprivileged instructions), it is no longer possible to completely isolate the guest OS.  3.3 Taxonomy of virtualization techniques 81 Guest In-memory representation Virtual Image Storage VMM Host emulation Virtual Machine Binary translation Instruction mapping Interpretation …… Host FIGURE 3.6 A hardware virtualization reference model. 3.3.1.2 Hardware-level virtualization Hardware-level virtualization is a virtualization technique that provides an abstract execution envi- ronment in terms of computer hardware on top of which a guest operating system can be run. In this model, the guest is represented by the operating system, the host by the physical computer hardware, the virtual machine by its emulation, and the virtual machine manager by the hypervisor (see Figure 3.6). The hypervisor is generally a program or a combination of software and hardware that allows the abstraction of the underlying physical hardware. Hardware-level virtualization is also called system virtualization, since it provides ISA to virtual machines, which is the representation of the hardware interface of a system. This is to differentiate it from process virtual machines, which expose ABI to virtual machines. Hypervisors A fundamental element of hardware virtualization is the hypervisor, or virtual machine manager (VMM). It recreates a hardware environment in which guest operating systems are installed. There are two major types of hypervisor: Type I and Type II (see Figure 3.7). Type I hypervisors run directly on top of the hardware. Therefore, they take the place of the operating systems and interact directly with the ISA interface exposed by the underlying hardware, and they emulate this interface in order to allow the management of guest operating systems. This type of hypervisor is also called a native virtual machine since it runs natively on hardware. 82 CHAPTER 3 Virtualization Type II hypervisors require the support of an operating system to provide virtualization services. This means that they are programs managed by the operating system, which interact with it through the ABI and emulate the ISA of virtual hardware for guest operating systems. This type of hypervisor is also called a hosted virtual machine since it is hosted within an operating system. VM VM VM VM ISA Virtual Machine Manager VM VM VM VM ABI ISA Operative System Virtual Machine Manager ISA ISA Hardware Hardware FIGURE 3.7 Hosted (left) and native (right) virtual machines. This figure provides a graphical representation of the two types of hypervisors. Conceptually, a virtual machine manager is internally organized as described in Figure 3.8. Three main modules, dispatcher, allocator, and interpreter, coordinate their activity in order to emulate the underlying hardware. The dispatcher constitutes the entry point of the monitor and reroutes the instructions issued by the virtual machine instance to one of the two other modules. The allocator is responsible for deciding the system resources to be provided to the VM: whenever a virtual machine tries to execute an instruction that results in changing the machine resources asso- ciated with that VM, the allocator is invoked by the dispatcher. The interpreter module consists of interpreter routines. These are executed whenever a virtual machine executes a privileged instruc- tion: a trap is triggered and the corresponding routine is executed. The design and architecture of a virtual machine manager, together with the underlying hardware design of the host machine, determine the full realization of hardware virtualization, where a guest operating system can be transparently executed on top of a VMM as though it were run on the underly- ing hardware. The criteria that need to be met by a virtual machine manager to efficiently support vir- tualization were established by Goldberg and Popek in 1974. Three properties have to be satisfied: Equivalence. A guest running under the control of a virtual machine manager should exhibit the same behavior as when it is executed directly on the physical host. Resource control. The virtual machine manager should be in complete control of virtualized resources.  3.3 Taxonomy of virtualization techniques 83 Virtual Machine Instance ISA Instructions (ISA) Dispatcher Interpreter Routines Allocator Virtual Machine Manager FIGURE 3.8 A hypervisor reference architecture. Efficiency. A statistically dominant fraction of the machine instructions should be executed without intervention from the virtual machine manager. The major factor that determines whether these properties are satisfied is represented by the lay- out of the ISA of the host running a virtual machine manager. Popek and Goldberg provided a clas- sification of the instruction set and proposed three theorems that define the properties that hardware instructions need to satisfy in order to efficiently support virtualization. THEOREM 3.1 For any conventional third-generation computer, a VMM may be constructed if the set of sensi- tive instructions for that computer is a subset of the set of privileged instructions. This theorem establishes that all the instructions that change the configuration of the system resources should generate a trap in user mode and be executed under the control of the virtual machine manager. This allows hypervisors to efficiently control only those instructions that would reveal the presence of an abstraction layer while executing all the rest of the instructions without considerable performance loss. The theorem always guarantees the resource control property when the hypervisor is in the most privileged mode (Ring 0). The nonprivileged instructions must be exe- cuted without the intervention of the hypervisor. The equivalence property also holds good since the output of the code is the same in both cases because the code is not changed. 84 CHAPTER 3 Virtualization Privileged Instructions Sensitive Instructions User Instructions FIGURE 3.9 A virtualizable computer (left) and a nonvirtualizable computer (right). THEOREM 3.2 A conventional third-generation computer is recursively virtualizable if: It is virtualizable and A VMM without any timing dependencies can be constructed for it. Recursive virtualization is the ability to run a virtual machine manager on top of another virtual machine manager. This allows nesting hypervisors as long as the capacity of the underly- ing resources can accommodate that. Virtualizable hardware is a prerequisite to recursive virtualization. THEOREM 3.3 A hybrid VMM may be constructed for any conventional third-generation machine in which the set of user-sensitive instructions is a subset of the set of privileged instructions. There is another term, hybrid virtual machine (HVM), which is less efficient than the virtual machine system. In the case of an HVM, more instructions are interpreted rather than being executed directly. All instructions in virtual supervisor mode are interpreted. Whenever there is an attempt to execute a behavior-sensitive or control-sensitive instruction, HVM controls the execution directly or gains the control via a trap. Here all sensitive instructions are caught by HVM that are simulated.  3.3 Taxonomy of virtualization techniques 85 This reference model represents what we generally consider classic virtualization—that is, the ability to execute a guest operating system in complete isolation. To a greater extent, hardware- level virtualization includes several strategies that differentiate from each other in terms of which kind of support is expected from the underlying hardware, what is actually abstracted from the host, and whether the guest should be modified or not. Hardware virtualization techniques Hardware-assisted virtualization. This term refers to a scenario in which the hardware provides architectural support for building a virtual machine manager able to run a guest operating system in complete isolation. This technique was originally introduced in the IBM System/370. At present, examples of hardware-assisted virtualization are the extensions to the x86-64 bit architecture introduced with Intel VT (formerly known as Vanderpool) and AMD V (formerly known as Pacifica). These exten- sions, which differ between the two vendors, are meant to reduce the performance penalties experienced by emulating x86 hardware with hypervisors. Before the introduction of hardware-assisted virtualiza- tion, software emulation of x86 hardware was significantly costly from the performance point of view. The reason for this is that by design the x86 architecture did not meet the formal requirements introduced by Popek and Goldberg, and early products were using binary translation to trap some sensitive instructions and provide an emulated version. Products such as VMware Virtual Platform, introduced in 1999 by VMware, which pioneered the field of x86 virtualization, were based on this technique. After 2006, Intel and AMD introduced processor extensions, and a wide range of virtualiza- tion solutions took advantage of them: Kernel-based Virtual Machine (KVM), VirtualBox, Xen, VMware, Hyper-V, Sun xVM, Parallels, and others. Full virtualization. Full virtualization refers to the ability to run a program, most likely an operating system, directly on top of a virtual machine and without any modification, as though it were run on the raw hardware. To make this possible, virtual machine managers are required to provide a complete emulation of the entire underlying hardware. The principal advantage of full virtualization is complete isolation, which leads to enhanced security, ease of emulation of different architectures, and coexistence of different systems on the same platform. Whereas it is a desired goal for many virtualization solutions, full virtualization poses important concerns related to perfor- mance and technical implementation. A key challenge is the interception of privileged instructions such as I/O instructions: Since they change the state of the resources exposed by the host, they have to be contained within the virtual machine manager. A simple solution to achieve full virtuali- zation is to provide a virtual environment for all the instructions, thus posing some limits on perfor- mance. A successful and efficient implementation of full virtualization is obtained with a combination of hardware and software, not allowing potentially harmful instructions to be executed directly on the host. This is what is accomplished through hardware-assisted virtualization. Paravirtualization. This is a not-transparent virtualization solution that allows implementing thin virtual machine managers. Paravirtualization techniques expose a software interface to the vir- tual machine that is slightly modified from the host and, as a consequence, guests need to be modi- fied. The aim of paravirtualization is to provide the capability to demand the execution of performance-critical operations directly on the host, thus preventing performance losses that would otherwise be experienced in managed execution. This allows a simpler implementation of virtual machine managers that have to simply transfer the execution of these operations, which were hard to virtualize, directly to the host. To take advantage of such an opportunity, guest operating systems 86 CHAPTER 3 Virtualization need to be modified and explicitly ported by remapping the performance-critical operations through the virtual machine software interface. This is possible when the source code of the operating sys- tem is available, and this is the reason that paravirtualization was mostly explored in the open- source and academic environment. Whereas this technique was initially applied in the IBM VM operating system families, the term paravirtualization was introduced in literature in the Denali project at the University of Washington. This technique has been successfully used by Xen for providing virtualization solutions for Linux-based operating systems specifically ported to run on Xen hypervisors. Operating systems that cannot be ported can still take advantage of paravirtualiza- tion by using ad hoc device drivers that remap the execution of critical instructions to the paravir- tualization APIs exposed by the hypervisor. Xen provides this solution for running Windows-based operating systems on x86 architectures. Other solutions using paravirtualization include VMWare, Parallels, and some solutions for embedded and real-time environments such as TRANGO, Wind River, and XtratuM. Partial virtualization. Partial virtualization provides a partial emulation of the underlying hard- ware, thus not allowing the complete execution of the guest operating system in complete isolation. Partial virtualization allows many applications to run transparently, but not all the features of the operating system can be supported, as happens with full virtualization. An example of partial vir- tualization is address space virtualization used in time-sharing systems; this allows multiple appli- cations and users to run concurrently in a separate memory space, but they still share the same hardware resources (disk, processor, and network). Historically, partial virtualization has been an important milestone for achieving full virtualization, and it was implemented on the experimental IBM M44/44X. Address space virtualization is a common feature of contemporary operating systems. Operating system-level virtualization Operating system-level virtualization offers the opportunity to create different and separated execu- tion environments for applications that are managed concurrently. Differently from hardware virtua- lization, there is no virtual machine manager or hypervisor, and the virtualization is done within a single operating system, where the OS kernel allows for multiple isolated user space instances. The kernel is also responsible for sharing the system resources among instances and for limiting the impact of instances on each other. A user space instance in general contains a proper view of the file system, which is completely isolated, and separate IP addresses, software configurations, and access to devices. Operating systems supporting this type of virtualization are general-purpose, time- shared operating systems with the capability to provide stronger namespace and resource isolation. This virtualization technique can be considered an evolution of the chroot mechanism in Unix systems. The chroot operation changes the file system root directory for a process and its children to a specific directory. As a result, the process and its children cannot have access to other por- tions of the file system than those accessible under the new root directory. Because Unix systems also expose devices as parts of the file system, by using this method it is possible to completely isolate a set of processes. Following the same principle, operating system-level virtualization aims to provide separated and multiple execution containers for running applications. Compared to hardware virtualization, this strategy imposes little or no overhead because applications directly use OS system calls and there is no need for emulation. There is no need to modify appli- cations to run them, nor to modify any specific hardware, as in the case of hardware-assisted  3.3 Taxonomy of virtualization techniques 87 virtualization. On the other hand, operating system-level virtualization does not expose the same flexibility of hardware virtualization, since all the user space instances must share the same oper- ating system. This technique is an efficient solution for server consolidation scenarios in which multiple application servers share the same technology: operating system, application server framework, and other components. When different servers are aggregated into one physical server, each server is run in a different user space, completely isolated from the others. Examples of operating system-level virtualizations are FreeBSD Jails, IBM Logical Partition (LPAR), SolarisZones and Containers, Parallels Virtuozzo Containers, OpenVZ, iCore Virtual Accounts, Free Virtual Private Server (FreeVPS), and others. The services offered by these technol- ogies differ, and most of them are available on Unix-based systems. Some of them, such as Solaris and OpenVZ, allow for different versions of the same operating system to operate concurrently. 3.3.1.3 Programming language-level virtualization Programming language-level virtualization is mostly used to achieve ease of deployment of applications, managed execution, and portability across different platforms and operating sys- tems. It consists of a virtual machine executing the byte code of a program, which is the result of the compilation process. Compilers implemented and used this technology to produce a binary format representing the machine code for an abstract architecture. The characteristics of this architecture vary from implementation to implementation. Generally these virtual machines con- stitute a simplification of the underlying hardware instruction set and provide some high-level instructions that map some of the features of the languages compiled for them. At runtime, the byte code can be either interpreted or compiled on the fly—or jitted5—against the underlying hardware instruction set. Programming language-level virtualization has a long trail in computer science history and orig- inally was used in 1966 for the implementation of Basic Combined Programming Language (BCPL), a language for writing compilers and one of the ancestors of the C programming language. Other important examples of the use of this technology have been the UCSD Pascal and Smalltalk. Virtual machine programming languages become popular again with Sun’s introduction of the Java platform in 1996. Originally created as a platform for developing Internet applications, Java became one of the technologies of choice for enterprise applications, and a large community of developers formed around it. The Java virtual machine was originally designed for the execution of programs written in the Java language, but other languages such as Python, Pascal, Groovy, and Ruby were made available. The ability to support multiple programming languages has been one of the key elements of the Common Language Infrastructure (CLI), which is the specification behind 5 The term jitted is an improper use of the just-in-time (JIT) acronym as a verb, which has now become common. It refers to a specific execution strategy in which the byte code of a method is compiled against the underlying machine code upon method call—that is, just in time. Initial implementations of programming-level virtualization were based on inter- pretation, which led to considerable slowdowns during execution. The advantage of just-in-time compilation is that the machine code that has been compiled can be reused for executing future calls to the same methods. Virtual machines that implement JIT compilation generally have a method cache that stores the code generated for each method and sim- ply look up this cache before triggering the compilation upon each method call. 88 CHAPTER 3 Virtualization.NET Framework. Currently, the Java platform and.NET Framework represent the most popular technologies for enterprise application development. Both Java and the CLI are stack-based virtual machines: The reference model of the abstract architecture is based on an execution stack that is used to perform operations. The byte code gen- erated by compilers for these architectures contains a set of instructions that load operands on the stack, perform some operations with them, and put the result on the stack. Additionally, specific instructions for invoking methods and managing objects and classes are included. Stack-based vir- tual machines possess the property of being easily interpreted and executed simply by lexical analysis and hence are easily portable over different architectures. An alternative solution is offered by register-based virtual machines, in which the reference model is based on registers. This kind of virtual machine is closer to the underlying architecture we use today. An example of a register-based virtual machine is Parrot, a programming-level virtual machine that was origi- nally designed to support the execution of PERL and then generalized to host the execution of dynamic languages. The main advantage of programming-level virtual machines, also called process virtual machines, is the ability to provide a uniform execution environment across different platforms. Programs compiled into byte code can be executed on any operating system and platform for which a virtual machine able to execute that code has been provided. From a development life- cycle point of view, this simplifies the development and deployment efforts since it is not neces- sary to provide different versions of the same code. The implementation of the virtual machine for different platforms is still a costly task, but it is done once and not for any application. Moreover, process virtual machines allow for more control over the execution of programs since they do not provide direct access to the memory. Security is another advantage of managed programming languages; by filtering the I/O operations, the process virtual machine can easily support sandboxing of applications. As an example, both Java and.NET provide an infrastructure for pluggable security policies and code access security frameworks. All these advantages come with a price: performance. Virtual machine programming languages generally expose an inferior performance compared to languages compiled against the real architecture. This performance dif- ference is getting smaller, and the high compute power available on average processors makes it even less important. Implementations of this model are also called high-level virtual machines, since high-level pro- gramming languages are compiled to a conceptual ISA, which is further interpreted or dynamically t