Power consumption has always been a challenging problem in data centers. Green data center has been proposed to alleviate the growing tendency of power consumption from the aspects of architecture and resource management. For improving the network performance and guaranteeing the quality of service (QoS) requirements of data flows, various kinds of network function virtualization (NFV) devices, which are equipped with multiple hardware resources, are ubiquitously deployed in data centers. Meanwhile, a chain of virtual network functions (VNFs) is usually needed to be sequentially executed on flows. Existing queueing schemes used in multi-resource environments are insufficient to guarantee the QoS requirements of flows in this situation. Based on this observation, we modeled and analyzed the flow scheduling problem in chains of VNFs. Further, we proposed a scheduling framework, named MRouting, to guarantee the deadline requirement of flows. MRouting determines the scheduling sequence of flows according to their urgency degree. It also implements appropriate data rate control and routing selection for flows, so that they can finish their transmissions before their deadlines. We have conducted extensive experiments to evaluate the performance of MRouting. The results verified that MRouting efficiently guaranteed the deadlines of flows, and also achieved obvious improvements on flow completion time and resource utilization. © 2018