Analyzing Bidding Strategies for Schedulers in a Simulated Multiple-Cluster Market Driven Environment
Publication Type: Paper
Repository URL: thesis_sindhura
Demand for high-end compute power has resulted in the emergence of the Computational Grid with several research systems facilitating its use. One such system Faucets, envisions the realization of Compute Power as a Utility, where the consumers can plug into the grid and tap required resources without being concerned about the source. For the efficient mapping of compute power producers and consumers in the system, Faucets employs the contract-net economic model. In this model, for each job consumers advertise their quality of service parameters and clusters reply with a bid. Bid value represents the amount a user is entrusted to pay in return for the successful completion of his job. Clusters participating in such a competing market need mechanisms for making profitable admission control decisions. The thesis presents a conservative deadline driven scheduling algorithm with lookahead for clusters called the GanttChart Strategy. This strategy guarantees completion of the accepted jobs within their respective deadlines in addition to ensuring efficient system utilization. In order to generate a bid value in a computational auction, clusters can use the GanttChart Strategy to aid in the private valuation of jobs. Further, we analyze the effect of dynamically changing the bid value depending on the cluster's utilization and present a model for attaining market equilibrium prices in the contract-net model of Faucets.
Sindhura Bandhakavi, "Analyzing bidding strategies for schedulers in a simulated multiple-cluster market driven environment", University of Illinois at Urbana-Champaign, 2003.