MRDC: Multicast Data Restoration in Fat-Tree Data Center Networks

Abstract

For data center networks (DCNs), it is very important to perform multicast transmission reliably. However, due to packet loss in DCNs, not only the reliability of data transmission should be guaranteed, but also the network throughput should be maintained stably. To this end, in this paper, we design a multicast fat-tree data restoration strategy (MRDC). The advantage of MRDC is that it can haul back to restore the server with data packet loss so that the server can receive the same data as the previous data transmission without the help of the source server. Thus, MRDC can effectively reduce the packet loss and improve transmission efficiency without increasing data restoration cost significantly, which is especially important for more reliable multicast. MRDC reduces subscribes in the network and avoid the congestion by choosing the optimal data restoration links. Simulation results show that MRDC can efficiently reduce packet loss rate, and increase throughput of servers while not increasing the link load and CPU usage rate significantly.