Research And Design Of An Energy-based Scheduler Used For Hybrid Cluster

With the risen of greenhouse gas emissions, global warming has become a problem which can not be ignored, and now more and more enterprises and research institutions have taken into account the issue of energy saving in the data center; With the number of servers increasing in the data center, the power used in the data center power is also increasing year by year, the problem of how to control the energy consumption in data center has become increasingly important.Currently, the main method to reduce energy consumption in the data center is virtualization. Interactive applications in the data center is typically deployed in a virtual cluster, but on the other hand, in order to avoid the performance overhead of virtualization, the batch jobs usually need to be deployed in the physical cluster.This paper presents an energy-based scheduler, which is used for hybrid cluster containing both the physical machine and virtual machines. In the premise of guaranteeing the performance, the energy-based scheduler can effectively reduce the energy consumption of the whole cluster. The scheduler scheduling process into two phases: in the first stage, the scheduler builds a model and predicts the run time of the task, and assign the tasks into a physical machine or a virtual machine based on the run time which has been predicted; in the second stage, the scheduler gets the information of each node status, monitoring and managing the energy consumption of each node.Finally, we built a hybrid cluster which included 8 physical servers and 24 virtual machines, and running the mixed load containing interactive applications and batch tasks. Experimental results show that hybrid cluster with the energy-based scheduler compared with virtual cluster, the task can be reduced the running time on average 33.4%, while reducing energy consumption by 12.1%; And compared with the physical cluster, although the loss of 14.6% performance,decreased 36.2 percent of energy consumption. Experimental results show that an efficient scheduling mechanism with hybrid data centers can provide a cost-effective solution for interactive and batch jobs.