Multilevel Exploration of the Optimization Landscape through Dynamical Fitness for Grid Scheduling

Joanna Kolodziej


Highly parameterized modern computational grids (CGs) are composed of large numbers of virtually connected various devices such as computers and databases. These systems must provide a wide range of services and should not be limited to high-performance computing platforms [1]. Typical grid users in one node or network cluster might not be able to have control over other parts of the system. Various types of information and data processed in the large-scale dynamic environment may be incomplete, imprecise, fragmentary, or overloading. All of those aspects make the scheduling and resource management problems in CGs challenging issues. Depending on the restrictions imposed by the grid application, different access policies in different network clusters, and different users'requirements, the complexity of those problems can be determined by the number of objectives to be optimized (single vs. multiobjective), the type of the environment (static vs.dynamic), the processing mode (immediate vs. batch), task interrelations (independence vs. dependency), and so on.

Theoretical analysis of the optimization landscapes for many classical combinatorial problems, such as the traveling salesman problem, graph bipartitioning, and flowshop scheduling [2], may be defined as a background of formal models of the wider NK family of the ...

Get Scalable Computing and Communications: Theory and Practice now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.