WRAPS Scheduling and Its Efficient Implementation on Network Processors

Network devices in high-speed networks need to support a large number of concurrent streams with different quality of service (QoS) requirements. This paper introduces a new packet scheduling algorithm and its efficient implementation on the novel programmable network processor-Intel’s IXP1200. WRAPS algorithm is based on the observation of packet rates within a dynamic window. The window can move continuously or discretely as the system transmits packets. Cumulated packet rates in the window are utilized to predict the next incoming packet of the same flow and reserve resource for the transmission of later packets. The implementation on network processor considers both accuracy and efficiency. To expedite the calculation and avoid the high cost of maintaining an ordered list, we designed a time- slotted circular queue to achieve O(1) insertion and selection time. Our experiments on the real system show good performance in terms of scalability and flow interference avoidance.