Resilient scheduling of cluster resources based on mixed load

Resilient scheduling of cluster resources based on mixed load

Li Yong Zhang Zhang mondain Han ji li Wang Yu

The current cluster resource scheduling method is difficult to adapt to the diversification of Internet business, the problem of customization feature, this paper puts forward a kind of cluster resource elastic scheduling method which is based on the mixed load. By constructing the job Constraint description language, this method allows the job to propose multidimensional resource request and load-aware resource scheduling algorithm on the basis of its own load characteristics. To achieve unified deployment and management of all types of business within the same cluster, to match the change of resource demand in time, through establishing the transformation mechanism between soft constraint and hard constraint to meet the requirement of customization of resources in different stages of operation. The experiment shows that this method can be used to achieve better performance with less resources than Hadoop. , in the actual production system, the utilization of cluster resources can be increased from 62% to 75% based on this method.

Resilient scheduling of cluster resources based on mixed load