A Research And Application Of Time-sharing Flow Scheduling For Datacenter Network

With the development of data center and generation of data center applications, different data center applications propose different performance requirements to network, which has led to many new data center technologies to be raised constantly. As a general application in a data center, web application becomes more and more complex after years of development. Render a web page may generate hundreds of internal network flow in the data center, which is usually short, typically tens of KB. In order to quickly respond to a user request, applications need to network flow with short transmission delay. While increasing data center network bandwidth can reduce the overall delay of the network to some extent, but when the bandwidth is increased to a certain extent, the delay of short data flows would not be improved. This is because there is a HOL blocking phenomenon in the data center network with mixed network flows. When the short flow enters the network after a long flow, this will lead to a short flow wait in the output queue of the switching device until the lone flow to be finished, which greatly increases the transmission delay of short flows. Based on these observations, in order to reduce the latency of network, especially the latency of short flows, we propose an algorithm of time-sharing scheduling based on priority queue. For the delay-sensitive short flows in the network, we assign high priority to them and put them into a high-priority output queue, while putting the lone flows into a low-priority output queue. When sending a packet, we send the data which is in the high priority queue first. To prevent long flows to be starved to death because of the high arrival rate of short flows, the algorithm would assign some time slices to the low-priority queue when it finds that the low-priority queue is not scheduled for a long time.The main contribution of this paper is to propose an algorithm of time-sharing scheduling based on priority queue to reduce the transmission delay of network, especially for the short flows. The algorithm proposed in this paper can significantly improve the real-time performance of the data center network, which provides better network services for the web applications in the data center.