Multi-Resource Scheduling with Consideration of Differential Services in Multi-Tenant Sharing

Current schedulers in the cloud computing platform are facing the performance requirement. However, due to lack of the theoretical and practical guidance on multi-resource scheduling, algorithms in the schedulers are neglecting the difference between resource demand vectors and resource demand flow, resulting into the large resource waste caused by fragmentation, with all tasks struggling with the same bottleneck resource. We refer the resource demand flow as the relationship between resource demand and time to build a more accurate task model from the perspective of resource consumption. Also, we present the offline sat scheduler, the reduction algorithm based on the sat solver, to give advice on the optimisation on scheduling order of the tasks with various resource demand flow. To make up for the time limit of the sat solver, we provide the price scheduler to optimise the resource utilisation, while promising a certain degree of fairness. The simulation shows that the performance of the sat scheduler beats YARN default scheduler by cutting 30% of the make span, profiting from SAT powerful calculating technique, and the price scheduler gets a shorter make span and higher utilization.