Grid Scheduling with Agents

Current efforts in Grid Computing operate on a zero settlement basis with users on the Internet downloading computation engines for “good causes” and subsequently returning their results. A number of attempts, for example Condor-G, GRaDS and NimrodG, have been made to develop computational economy based scheduling systems. However, these systems operate on a “top down” basis and have a number of limitations, in that they assume knowledge of all the jobs on the grid, they require the ability to manage the scheduling of jobs on grid nodes and they assume a direct linkage between the “grid to node” and “within node” scheduling processes. Our work in the area is focused on examining how we may use multi-agent systems to develop a market place for the trading of computational grid resources such that grid scheduling is completed without the limitations of the “top down” approach.