Knowledge discovery for scheduling in computational grids
Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery
Improving scheduling performance using a q-learning-based leasing policy for clouds
Euro-Par'12 Proceedings of the 18th international conference on Parallel Processing
| Hi-index | 0.00 |
Grid scheduling is a key problem for Grid to improve the resource management and application performance. It has been proven to be a NP-hard problem for the computation of optimal Grid schedules, which is responsible to allocate resources to user jobs with the objective such as minimizing the completion time or cost. Therefore, it is more difficult for Grid scheduling system to cope with the dynamically varied resource and jobs. To solve this problem, an adaptive negotiation based scheduling model is presented. The near-optimal schedules are selected by learning agents representing the resource and jobs respectively in Grid. The agents can reduce the size of scheduling search space through a modified reinforcement learning algorithm, where the state-value function is improved by a numerical function approximation and the balance of efficiency and complexity is obtained by a simulated annealing algorithm. The results demonstrate that the proposed negotiation model and the learning agents based negotiation model are suitable and effective for Grid environments.