Research Of Grid Task Scheduling Algorithms On Computational Economy

In recent years, with the rapid development of the high-speed networks, it ispossible to integrate the distributed and heterogeneous computing resources intoComputational Grid. It provides an ideal platform to solve large-scale computingproblems in scientific and engineering area. To let the Grid be the mainstream in thecomputational model, we have to develop an effective cure to make the users put outtheir resources whose are idled. And introduce market economy concept to the gridmanagement and design is a good choice.Deadline and Budget Constrained is based on the most important factorsincluding time and cost of the task scheduling. There are three heuristic algorithms inthe policy, each has its target to optimize including time and cost.Time Optimization is one of the DBC Scheduling Policy aiming to minimize thecomputational time of task scheduling. There are two problems in its scheduling, oneis the load imbalance, the other is that the number of completed tasks is small whenthere is no enough budgets. We propose a method that changes the assignment of thetasks and removes excess tasks from the task aggregate by forecasting the time andcost of the last tasks. Finally, the result of simulation experiment shows that ourheuristic shortened the time and the ratio of succeeded tasks remain very high whenthere is no enough budgets.Aim to the fact that the three heuristic algorithms in the DBC Scheduling Policyonly can optimize one target of the two QoS targets, we proposed performance/priceheuristic algorithm. From the user point, it uses the remaining time of the taskscheduling as the performance and use the cost of the task scheduling as the price todistribute resources for the tasks. It considers both the time and cost of taskscheduling. The result of simulation experiment shows when the deadline is low, usercan get the service with a high quality. When the difference of the cost between services is not so much, the less time task scheduling does, the more user satisfies. The performance/price heuristic algorithm is an efficient algorithm that can provide one new choice for user.