Research On Data Center Resource Scheduling And Storage Performance Optimization Based On Mesos

In order to meeting the data requirements for efficient management and maintenance of enterprise,the data center virtualize the physical server resources of the data center.It also implements unified management and scheduling of data center resources.Mesos is an important component of the current data center management system.Therefore,in order to effectively improving the performance of Mesos in data center practice,this thesis focuses on analyzing the key technologies of Mesos resource scheduling and persistent storage,improving and optimizing them separately.(1)DRF algorithm is a default fair scheduling algorithm of Mesos,which does not consider the classification of the machine performance and task type.This thesis proposes an improved DRF fair scheduling algorithm X-DRF pre-aware task type.The algorithm can not only ensure the fairness of resource allocation,but also make the system more reasonable allocation of resources.It would further improve the system's resource utilization.Finally,the effectiveness of the method is proved by experimental comparison.It provides a new idea for data center managers to optimize the data center performance based on the Mesos resource scheduling system.(2)This thesis analyzes Mesos persistent storage methods.It proposes a Mesos persistent storage optimization method for HDD-SSD based on the characteristics of Mesos persistence.Firstly,the method uses HHD-SSD hybrid storage HDFS as a Mesos persistence layer.Then,for the problem of I/O bottleneck caused by frequent read and write operations,the method uses a memory cold page based strategy for the HDFS.Finally,the experimental comparison shows that this method can not only improve the I/O performance of the data center based on the Mesos resource scheduling system,but also ensure the utilization of the SSD.