heterogeneity. A cloud must support the aggregation of heterogeneous hardware and software resources, as it happens with scientific experiments. The concept of virtualization [2] is also a key aspect for clouds.

Through virtualization, many users may benefit from the same infrastructure using independent instances. Virtualization enables the first security [10] level in the clouds, since it allows the isolation of environments. In clouds, each user has unique access to its individual virtualized environment. Resource sharing is provided by clouds, since each resource is represented as a single artifact, giving the impression of a single dedicated resource. Scalability is mainly defined by increasing the number of working nodes. By definition, clouds offer the automatic resizing of virtualized hardware resources. Monitoring refers to the ability of watching the current status of virtual machines or services provided.

Each one of those architectural characteristics is standardized by specific standards (which are in another class of the taxonomy). Besides that, some architectural characteristics are important to scientific experiments, such as scalability and monitoring to control the execution.

3.3.5  Technology Infrastructure

The technological infrastructure subtaxonomy (Fig. 3.7) is responsible to classify a cloud environment according to the computational power provided by cloud environments. Particularly in commercial clouds, scientists have no access to the kind of technology that is used to implement it. In fact, in commercial cloud environments implementation details are hidden from the end-user (scientist). On the other hand, in academic or private clouds it is possible to obtain this information. This information may be quite useful in e-Science because many experiments need a powerful computational environment to run and if the cloud environment is not able to provide powerful resources, it will not be able to support these experiments. But, it is complicated to scientists to choose between those environments (clusters, blades, or grids) to run experiments, since they may not be computer experts. This way, we need to classify the environments using another classification.

Fig. 3.6  Architecture subtaxonomy

Fig. 3.7  Technology infrastructure subtaxonomy

Fig. 3.9  Standards subtaxonomy

Fig. 3.10  Orientation subtaxonomy

API authorization in a simple and standard way. On the other hand, OpenID [21] is an open, decentralized standard for user authentication and access control, allowing users to log onto many services with the same digital identity, as adopted on grids. In addition, we may find SAML [17], which is a major player in cloudbased systems. Atom Publishing Protocol [3] (or simply Atom) is a content licensing protocol based on HTTP for creating and updating web resources. RSS [23] must be included as a syndication standard as well. Even RSS is not a recommended standard but a de facto standard. As highlighted on the scientific experiments requirements, security is a key aspect and virtualization improves security. Virtualization is a key aspect of cloud computing and needs some standards. The OVF [22] is being considered as one of the de facto standards for virtualization. OVF enables flexible and secure distribution of software and data, facilitating the mobility of virtual machines. As happens in many web systems, data is usually represented and transferred using XML (and many more XML-based languages such as SAML [17], XACML [32], and JSON [11]). JSON is a lightweight datainterchange format.

3.3.8  Orientation

One important aspect of cloud computing for e-Science is the orientation (taxonomy represented in Fig. 3.10). Usually, the orientation changes as the type of service changes. For instance, when an application is provided on the cloud, we may consider it task-centric, because it is oriented to the task that will be executed. In other words, you need to transfer control to the application owners instead of

having­ control of it. However, when the infrastructure is provided as a service, the user has control of the process. The programs, applications, and data are chosen by the user. Thus, the cloud may be user-centric.

3.4  Classifying Cloud Computing Environments

Using the Taxonomy

In this section, we present a summarized survey of the main existing cloud computing environments according to the proposed taxonomy. Table 3.1 shows the selected cloud computing environments with their categorization based on the taxonomy. These cloud computing environments are the most commonly found in scientific literature [9, 14, 15, 24].

In Table 3.1, we may observe that none of the analyzed environments provides all functionalities and characteristics presented in the proposed taxonomy. Scientists will have to analyze their needs and verify in the classification the environment that is the most suitable. For example, suppose that scientific experiments require HPC support, API, and privacy as its main requirements. In a first analysis, scientists would choose between Nimbus and Eucalyptus. However, if a database service is also an important issue to be considered, they might trade between the available environments.

Table 3.1  Classification of cloud computing environments

a http://aws.amazon.com/ec2/ b  http://www.microsoft.com/windowsazure/ c www.nimbusproject.org/nimbus_cloud d http://open.eucalyptus.com/ e http://www.ibm.com/ibm/cloud/

f Subject to acceptance g  Free for tests

h Microsoft Azure is composed by Windows Azure, Microsoft SWL Azure, and Windows Azure platform AppFabric

i Information not available