Traffic in data centers mainly consists of elephant flows and mice flows, but elephant flows are primarily responsible for network congestion because they usually carry large amount of data. To avoid network congestion and balance network load, many recent works have proposed dynamic and centralized flow scheduling approaches based on SDN which can control the forwarding of elephant flows with the help of a centralized controller. However, in the existing approach Ashman, the controller queries switches periodically with a static polling period, which cannot adapt to traffic dynamics and may cause a great monitoring overhead of the controller. Moreover, the network throughput of Ashman can be further improved. In this paper, we present EAshman, a low-cost flow scheduling framework that focuses on enhancing the performance of Ashman. To reduce the overhead of the controller, EAshman only reschedules elephant flows that go through congested links and an adaptive polling period adjustment algorithm is proposed to dynamically adjust the polling period based on the current network load. Besides, we also propose a Probability-based path selection algorithm to improve network throughput. Simulation results show that EAshman can significantly save the overhead of the controller and achieve higher network throughput compared to Ashman