An agent-based network-aware approach to grid resource scheduling

Grid applications are becoming increasingly network dependent with more demanding requirements in areas such as data access or interactivity, and consequently the quality and consistency of the underlying transport network services are among the most important factors in scheduling policies. The Grid scheduling system has the responsibility to distribute the demands in the best possible way, taking into account application profiles and resource availability. However, traditional Grid scheduling has been based on a partially or totally network unaware approach, where the selection of computational resources is forced to be co-located with that of storage, because without any information on network connectivity, it is senseless to allocate the computation far from the storage. Therefore, what is needed is a novel meta-scheduling and Resource Management Architecture for large data Grids to deal with the co-allocation problem where applications have resource requirements that can be satisfied only by using resources simultaneously at several sites. An active network-aware approach to Grid scheduling needs a strict integration between the scheduling logic, the requesting application and the network entities that offer the connectivity services. In our approach, the Grid has control of the network, as it is a local resource similar to other local resources. On the other hand, the meta-scheduling process must be scalable, fault-tolerant, flexible and self-organizing enough to be able to efficiently allocate data-intensive distributed task in such a way that the impact of network performance is minimized. Our approach is based on encapsulating all the Grid control logic into mobile agents. Accordingly, we conceived a new network-aware Grid resource scheduling approach based on a combination of both intelligent agents and multi-agent facilities and autonomic self-organization strategies, to meet the requirements of smart self-management and network efficiency of the next generation intelligent grid environments. The agent flexibility can greatly improve the reservation acceptance rate on the Grid. The improvement is even more important when the meta-scheduler tries to co-reserve multiple resources from independent sparse resource managers for the same period of time. © 2009 Nova Science Publishers, Inc. All rights reserved.