Resource Allocation for Cloud Computing: A Semantic Approach

Introduction

In the cloud ecosystem (Vaquero, Rodero-Merino, Caceres, & Linder, 2008), we can find different providers which offer computing Infrastructure as a Service (IaaS) on demand. This group of providers and their resources has been called Public Cloud. With the evolution of virtualization technologies, enterprises start to organize their big datacenters as Private Clouds dedicated for deploying services and executing computational tasks which are required by their business. This virtualized organization of datacenters facilitates the management and configuration of the computing infrastructure, automating repetitive processes as well as allowing the outsourcing of computation to Public Clouds in peaks of computing demand. However, the outsourcing of computation is not an easy task: it has to coordinate different actions and raises several problems. The different Cloud Providers offer heterogeneous resources described according to their own metrics or properties which are different in other providers. Those descriptions must be understood by the customers and compared to their requirements. Once the descriptions have been understood, the providers’ resources must be allocated to customers’ tasks taking into account not only the technical requirements but also the customers and providers rules. After the resource allocation, the customers’ tasks must be executed on the selected resource and, finally, the execution must be monitored for detecting unexpected events adapting the execution for solving them.

This chapter presents the Semantically Enhanced Resource Allocator (SERA), a toolkit for allocating customers’ tasks to resources from different resource providers benefitting from the semantic web (Berners-Lee, Hendler, & Lassila, 2001) and multi-agent (Ferber, 1999) technologies. On one hand, semantic technologies are used to achieve semantic interoperability between resource descriptions from different providers and different customers’ requests. On the other hand, agents increase the autonomy and self-management of a system. Agents are proactive, so, they can take decisions by themselves according to their beliefs, desires and intentions and trigger actions in the system by their own initiative. For this reason, agents are suitable for coordinating the allocation and execution of tasks in these clouds environments, detecting execution problems and reacting to them by triggering the most appropriate actions depending on the system status. They are also capable to communicate each other implementing negotiation protocols which are very useful to reach agreements with other agents. Benefiting from this capability, the resource allocation in our methodology is agreed by means of a negotiation between the customer and provider agents taking into account the customers and providers rules.

The capabilities presented above enable the SERA toolkit to support different customers’ requests (such as batch jobs and service deployments) and to interoperate with different computing platforms (such as Grids and Clouds) giving a general resource allocation solution. For this reason, some SERA parts (such as components, agents and ontology classes) are presented in general. However, in the different chapter sections we have tried to give details about how these concepts specifically apply to the Cloud Computing environment. The chapter is organized as follows: first section presents the SERA toolkit architecture, focusing in the most important concepts: the ontology for resource allocation in clouds, the automatic annotation of resource descriptions from the different providers' schemes and the multi-agent based semantic resource allocation process. Following, we provide installation and deployment guidelines as well as different examples on how to use the SERA toolkit. The chapter is finalized drawing conclusions about the lessons learned and presenting future research directions.