Joint Resource Allocation for Online and Offline Tasks in Data Centers

Nowadays an increasing number of online tasks and offline tasks from all industries are accommodated by the widely deployed data centers. These online tasks usually have strict Quality of Service (QoS) requirements and follow periodic load patterns. It is important to co-locate online tasks and the offline tasks that are not so sensitive to time delays to improve the resource utilization of data centers. Most existing solutions only focused on the resource management for online tasks, and ignored the features of offline tasks. In this paper, we establish the performance models for online tasks and offline tasks, respective-ly, and propose a virtual computing resource allocation scheme based on the dynamic resource sharing strategy. This scheme utilizes the time-varying characteristics of online workloads to maximize the utilization of the temporary available resource, while guaranteeing the QoS of delay-sensitive services. Thus the impact of hybrid deployment on the execution of offline tasks can be alleviated. The simulation results showed the effectiveness of the proposed scheme in guaranteeing the QoS and improving the resource utilization in data centers.