Resource Management

Resource Management

Valentin Cristea (Politehnica University of Bucharest, Romania), Ciprian Dobre (Politehnica University of Bucharest, Romania), Corina Stratan (Politehnica University of Bucharest, Romania) and Florin Pop (Politehnica University of Bucharest, Romania)
DOI: 10.4018/978-1-61520-703-9.ch004

Abstract

The resource management is an important component in LSDS implemented for a variety of architectures and services. This chapter considers the management of distributed resources, virtual resources and provides the requirements for resource management in large scale distributed system. A resource management system is defined as a service that is provided by a distributed network component system that manages a pool of named resources that is available such that a system- centric or job-centric performance metric is optimized. Due to issues such as extensibility, adaptability, site autonomy, QoS, and co-allocation, resource management systems is more challenging in large scale distributed computing environments. The taxonomy of resource management systems (RMS) for very large-scale network computing systems presents the variety of requirements for this tool. The taxonomy could be used to identify architectural approaches and issues that have not been fully explored in the research. The resource management system could support different users constrains, so the multiple policies is provided. In general, requiring the RMS to support multiple policies can compel the scheduling mechanisms to solve a multi-criteria optimization problem. An important subject presented in this chapter is Agents Frameworks for resource management that offer a mechanism for distributes resources management. The chapter ends with presentation of WSRF (Web Services Resource Framework) that is the new solution for resources management based on SOA (OGSA – Open Grid Service Architecture). Resource management in Grid implies a quite large number of functionalities, from resource discovery to scheduling, execution management, status monitoring and accounting. In this section, we shall focus on scheduling systems, and we shall present the monitoring functionalities and the Grid information systems in a further section. We shall introduce here some general issues, and then we shall present taxonomy of the scheduling systems and some details regarding the scheduling mechanisms used in the most important current Grid projects.
Chapter Preview
Top

Introduction

The resource management is an important component in LSDS implemented for a variety of architectures and services. This chapter considers the management of distributed resources, virtual resources and provides the requirements for resource management in large scale distributed system. A resource management system is defined as a service that is provided by a distributed network component system that manages a pool of named resources that is available such that a system- centric or job-centric performance metric is optimized. Due to issues such as extensibility, adaptability, site autonomy, QoS, and co-allocation, resource management systems is more challenging in large scale distributed computing environments. The taxonomy of resource management systems (RMSs) for very large-scale network computing systems presents the variety of requirements for this tool. The taxonomy could be used to identify architectural approaches and issues that have not been fully explored in the research.

The resource management system could support different users constrains, so the multiple policies is provided. In general, requiring the RMS to support multiple policies can compel the scheduling mechanisms to solve a multi-criteria optimization problem. An important subject presented in this chapter is Agents Frameworks for resource management that offer a mechanism for distributes resources management. The chapter ends with presentation of WSRF (Web Services Resource Framework) that is the new solution for resources management based on SOA (OGSA – Open Grid Service Architecture). Resource management in Grid implies a quite large number of functionalities, from resource discovery to scheduling, execution management, status monitoring and accounting. In this section, we shall focus on scheduling systems, and we shall present the monitoring functionalities and the Grid information systems in a further section. We shall introduce here some general issues, and then we shall present taxonomy of the scheduling systems and some details regarding the scheduling mechanisms used in the most important current Grid projects.

Wieder et al. (2006) distinguishes between two cases of Grid systems with respect to their requirements on resource management capabilities.

Case 1 is Specialized Grids for dedicated purposes, which are centered on a single or limited application domain and require high efficiency in execution. The Resource Management System (RMS) is adapted to the special application, its workflow and the available resource. Thus, the interfaces to the resources and the middleware are built according to the given requirements caused by the application scenario. While the Grid RMS is highly specialized, the handling for the user is often easier as the know-how of the application domain has been built into the system.

Case 2 is a Generic Grid Middleware, which has to cope with the complete set of the requirements to support applicability. The Grid RMS is open for many different application scenarios. In comparison to the specialized Grids, generic interfaces are required that can be adapted to many front- and backend. However, the generic nature of this approach comes at the price of additionally overhead for providing information about the application. For instance, more information about a particular job has to be provided to the middleware, such as a workflow description, scheduling objectives, policies and constraints.

The application-specific knowledge cannot be built into the middleware, and therefore must be provided at the frontend level. In this case, the consideration of security requirements is an integral aspect, which is more difficult to solve. It is possible to hide the additional RMS complexity of generic Grid infrastructures from the users or their applications by specialized components, which might be built on top of a generic middleware. Nevertheless, it can be concluded that in general a generic Grid middleware will carry additional overhead with less efficiency at the expense of broader applicability.

Top

Requirements For Resource Management In Lsds

Current research is mostly focusing on presented Case 1 in which solutions are built for a dedicated Grid scenario in mind. As mentioned before, these systems are usually more efficient and will therefore remain the favorite solution for many application domains. That is, Case 1 will not become obsolete if corresponding requirements and conditions exist. However, for creating future generation Grids suitable solutions are required for Case 2.

Complete Chapter List

Search this Book:
Reset