Geetha Manjunath

Platform as a Service

Platform as a Service is a cloud service model where the vendor provides a platform for development and deployment of cloud application over an abstracted hardware. PaaS solutions enable users to directly develop their applications without worrying about the complexity of setting up the hardware or system software. This chapter describes some popular cloud solutions that are examples of the Platform as a Service (PaaS) model of cloud delivery. The platform case studies discussed in this chapter are Microsoft Azure, Apache Hadoop, IBM PureXML, Yahoo! Mashups and others. Azure is a popular platform that enables developers familiar with Windows-based programming to create cloud applications using .NET. Hadoop is yet another cloud platform that has received a lot of developer attention due to the simplicity with which it enables one to create high-performing distributed applications. Platforms that enable data-oriented applications such as Mashups and those that provide special features for data manipulation such as pureXML from IBM are discussed in this chapter as well.

Combining heterogeneous classifiers for relational databases

Practical usage of machine learning is gaining strategic importance in enterprises looking for business intelligence. However, most enterprise data is distributed in multiple relational databases with expert-designed schema. Using traditional single-table machine learning techniques over such data not only incur a computational penalty for converting to a flat form (mega-join), even the human-specified semantic information present in the relations is lost. In this paper, we present a practical, two-phase hierarchical meta-classification algorithm for relational databases with a semantic divide and conquer approach. We propose a recursive, prediction aggregation technique over heterogeneous classifiers applied on individual database tables. The proposed algorithm was evaluated on three diverse datasets, namely TPCH, PKDD and UCI benchmarks and showed considerable reduction in classification time without any loss of prediction accuracy.

A Practical Heterogeneous Classifier for Relational Databases

Most enterprise data is distributed in multiple relational databases with expert-designed schema. Using traditional single-table machine learning techniques over such data not only incur a computational penalty for converting to a ”flat” form (mega-join), even the human-specified semantic information present in the relations is lost. In this paper, we present a two-phase hierarchical meta-classification algorithm for relational databases with a semantic divide and conquer approach. We propose a recursive, prediction aggregation technique over heterogeneous classifiers applied on individual database tables. A preliminary evaluation on TPCH and UCI benchmarks shows reduced training time without any loss of prediction accuracy.

Future Trends and Research Directions

This chapter describes future trends and research directions in cloud computing. The first few sections describe upcoming standards for cloud platform vendors and benchmarks that enable a cloud user to compare multiple cloud platforms. OpenCirrus, an open cloud research testbed, is described along with some of the details of novel tools developed for cloud-scale management. Since this book is developer oriented, the future of cloud application development leading towards end user programming is also covered in this chapter. Finally, the chapter concludes with a list of open research problems in cloud computing.

Infrastructure as a Service

This chapter describes an important cloud service model called “Infrastructure as a Service” type (IaaS), that enables computing and storage resources to be delivered as a service. The chapter takes popular cloud platforms as case studies, describes their key features and programming APIs with examples. To provide an insight into the trade-offs that the developer can make to effectively use the system, the chapter also contains a high level description of the technology behind the platforms. A more detailed internal systems view of the technology challenges and possible approaches to solve them are detailed in Chapter 6 .

Designing Cloud Security

One of the main hurdles for adoption of Cloud computing by public and enterprises is that of security and privacy of data hosted on the cloud. While the technical solution for the same can actually use existing security protocols for web services, the key problems to be addressed are in processes and governance. This chapter describes the processes and best practices that need to be implemented for a secure cloud. Additionally, it also describes the security issues introduced by a public cloud service provider. This is an abridged version of the book Securing the Cloud by Vic (J.R.) Winkler.

Managing the Cloud

As studied in earlier chapters, cloud platforms need to ensure that the system supports on-demand scaling as well as enables a pay-per-use model. Both of the above needs require that the system be monitored on a fine time scale. Additionally, management of cloud infrastructure is nontrivial because of the scale of cloud systems with multiple users. Challenges arise in every aspect of management, such as allocating resources in such a way as to ensure that SLAs are not violated; troubleshooting performance problems and failures; or upgrading the cloud infrastructure. This chapter describes approaches to these problems for IaaS, PaaS and SaaS systems.

Paradigms for Developing Cloud Applications

This chapter provides some deep insights into the new design paradigms that have evolved for cloud computing applications. It describes the fundamental concepts and theoretical knowledge that enable developers to develop efficient cloud applications on these new platforms. Specifically, the technologies for enabling scalable data storage using database partitioning and key-value stores are described. Also, a deep description of the MapReduce programming paradigm that is needed to break a compute-intensive task into the Map-Reduce workflow is provided. It also describes ways of developing client applications that provide a rich and immersive experience despite being distributed between the client device and the cloud.

Addressing the Cloud Challenges

Publisher Summary This chapter describes some approaches to address the key challenges posed by cloud computing. The scalability challenges needs to be addressed both at the computational as well as the storage access level. The chapter starts with architectures for linearly scaling the compute capacity by just adding more servers. For compute scalability, if the application instances are independently scheduled on independent processors, then just adding more servers can scale to a large number of clients. Amdahls Law puts bounds on the amount by which an application can be scaled. For 3-tier applications, there are two ways of scaling, by either adding compute power to each of the tiers of the application or by breaking down the application components in such a way that the individual components (application nodes) are self-sufficient with adequate amounts of database and messaging bus. Following this, it discusses key theoretical concepts that help a developer to appreciate the performance bottlenecks that may arise in an application due to concurrent data access and solutions that one can employ to address them. Further, solutions and approaches to address the issue of enabling multi-tenancy and architectures to ensure highly available cloud-hosted applications are explored.

Chapter 9 – Related Technologies

Publisher Summary This chapter reviews multiple technologies that are often confused with cloud computing, such as Grid computing, Utility computing, Distributed computing, and Virtualization. Each such technology is briefly described and the similarities, differences, and ways of leveraging it within a cloud infrastructure are discussed. The chapter also looks at architectures for network-based block virtualization where the abstraction could be provided either by having smart routers or on specialized appliances, through case studies from popular SAN solutions from HP and IBM. Virtualization is a very important and fundamental technology that enables hardware resources to expand or contract in an application transparent manner. Virtualization enables some of the key characteristics of a cloud infrastructure. Understanding and appreciating the complexities in implementing these techniques enables a developer to look at cloud infrastructure as a holistic solution. Grid computing is considered next where its similarities and differences with respect to cloud computing are studied. Finally, other cloud-related terminologies are briefly discussed.