Global behavior modelinga new approach to grid autonomic management

Abstract

Over the last decade, grid computing has paved the way for a new level of large scale distributed systems. The grid can be defined as a set of geographically dispersed, interconnected computational resources, aimed at performing challenging computational tasks. This infrastructure makes it possible to securely and reliably take advantage of widely separated computational resources that are part of several different organizations. Resources can be incorporated to the grid, building a theoretical virtual supercomputer. However, this new step in distributed computing comes along with a completely new level of complexity. Grid management mechanisms play a key role, and a correct analysis and understanding of the grid behavior is needed. Grid systems must be able to self-manage, incorporating autonomic features capable of controlling and optimizing all grid resources and services. Traditional distributed computing management mechanisms analyze each resource separately and adjust specific parameters of each one of them. When trying to adapt the same procedures to grid computing, the vast complexity of the system can make this task extremely complicated. But grid complexity could only be a matter of perspective. It could be possible to understand the grid behavior as a single system, instead of a set of resources. This abstraction could provide a deeper understanding of the system, describing large scale behavior and global events that probably would not be detected analyzing each resource separately. This abstraction could also be a solid, unified basis on top of which advanced grid autonomic management solutions could be developed. In this Ph.D. thesis a specific methodology is presented and described in order to create a global behavior model of the grid, analyzing it as a single entity. The purpose of this model is to serve as the above mentioned abstraction of the grid system, providing an unique global behavior understanding. This global behavior model becomes also an extremely valuable tool for developing autonomic management mechanisms and contributes to a service-oriented, unified vision of the grid. This methodology is strongly based on system monitoring, performance estimation tools, knowledge discovery techniques and advanced scientific visualization. As a whole, it provides a unique and new point of view of grid computing, contributing to enrich and develop this technology. To conclude, the proposed methodology has been tested on a series of typical experimental scenarios, both real and simulated, obtaining statistically meaningful results confirming that a global behavior model of a grid system benefits its understanding and serves as a solid basis to improve its autonomic capabilities.