A Decentralized Recent Neighbor Load Balancing Algorithm for Computational Grid

The computational grid is a distributed paradigm that providesresources for large scientific applications. It typically consistsof heterogeneous resources such as clusters that may reside indifferent administrative domains and be connected bynetworks with widely varying performance characteristics.Many scheduling and load balancing solutions have beenproposed for such computational grids. However, they allsuffer from significant deficiencies when extended to aheterogeneous grid environment. Computational grids havethe potential for solving large scale scientific computingapplications. The main techniques that are more suitable tocope with the dynamic nature of the grid are the effectiveusage of the grid resources and the distribution of theapplication load among multiple resources in the gridenvironment. In this paper, a decentralized grid model, as acollection of clusters has been proposed. The problem of loadbalancing in grid architecture is addressed by assigning loadsin a grid without neglecting the communication overhead incollecting the load information.An efficient dynamic load balancing algorithm named as‘Recent Neighbor’ (RN) has been presented to tackle theabove new challenges. RN performs intra-cluster and intercluster(grid) load balancing.The simulation results show that the proposed ‘RecentNeighbor’ algorithm for heterogeneous grid is feasible andimproves the system performance considerably.