A market-based scheduling algorithm for the allocation of resources to tasks in grid computing is proposed in this chapter. The method has been implemented in the context of resources represented by common object request broker architecture (CORBA) objects. Users specify not only the estimated runtime of the task, early start, late finish, and so on, but also the maximum number of hops to forward their requests, and other parameters. In addition, users specify a utility function, which is used to compute the utility value of the offers for resources. Each site meta-manager has an objective function used to find the equilibrium between supply and demand.
There are a number of details in the chapter that are missing or unclear. For instance, the text says “this process is an auction with neither a centralized nor a decentralized auctioneer.” However, the paper fails to say how the equilibrium is reached. There seems to be an implicit assumption that every task can be scheduled, either locally or remotely. Nothing is said of what happens when this is not the case.
Some experimental results are shown. They are limited to a single multiprocessing system, with no multi-site scheduling. The experimental evaluation is limited, and very little is said about the effects of different objective functions and different user utility functions. Multi-site scheduling is possible, but it is not described with enough detail. Better figure captions would have made reading the chapter easier. For instance, the caption for figure 30.4 is uninformative, and the caption for figure 30.7 fails to define the acronym AWRT, used on the y-axis. In conclusion, the chapter touches on an interesting direction for grid scheduling, but its lack of detail reduces its value.