Abstract:Resource scheduling is a key technique for container management. Prior work or meets the goal of fairness, the work load average is scheduled to all physical nodes, pays attention to the throughput indicator; or to meets performance targets, multiple carrier scheduling the workload related to physical nodes in the same or similar, pay attention to the response time. In this paper, it presents an application-aware resource scheduling approach, and employs a multi-queue model to meet both fairness and performance targets. Experimental results show that the approach has equal throughput for typical big data processing scenarios, and can reduce latency by up to 100% for typical transactional application scenarios.