The Grid technology is transforming the form of carrying out computation, communication, interconnection and the solution of large scale scientists problems, engineering and business. Currently, there are numerous scientific and business projects that make use of Grid technology with successful results. Such is the case of the LCG, The Large Hadron Collider (LHC) Computing Grid, at CERN, the European organization for nuclear research. It is the proliferation of independent Grids systems which has brought to light the need for federal structures allowing integration and sustainable resource control. The most representative example of this need it is the “European Grid Initiative” (EGI). Although a Federated Grid can consist of several different types of Grid infrastructures, is still based on the same principle that any Grid system, namely in coordinating resources that are not subject to centralized control.

Scheduling problem is one of the best known in Computer Science, however, applying existing scheduling algorithms to a Federated Grid environment presents several problems, mainly for scalability. So much so that with the emergence of such systems the trends in scheduling have undergone a change of address from local to global scheduling. The main reason why you can not take advantage of previous research is because the assumptions that are the basis of centralized systems are not applicable in a Grid environment. Therefore, scheduling strategies based on these ideas produce bad results in practice. As a result, one of the most important objectives of this PhD is to design a Federated Grid decentralized architecture based
on meta-schedulers.

Unlike the local manager, the meta-scheduler has an overview of the entire Federated Grid. Therefore, fine-grain scheduling policies are not appropriate for this level. These techniques are better suited for local managers, since these completely control the resources found in the layers closest to them. In contrast, the metascheduler needs light, decoupled, and coarse-grain techniques. In this sense, the main objective of this PhD is the study and analysis of various scheduling algorithms that follow these principles, based on a performance model, enabling the scheduling of independent jobs in Federated Grids and also able to reduce the makespan of applications and increase performance of resources.

The main advantage of using a performance model on which to base our mapping strategies lies in the lack of dependence on resources state. In this respect, we find solutions, also at the meta-scheduler level, who use information on the state of resources. However, it is well known that centralized and hierarchical information services present significant limitations, such as single point of failure, lack of scalability and high cost in bandwidth.