Applying LSTM to Enable Cache Prefetching to Optimize Flow Table Update Efficiency in SDN Switches

Abstract

Due to the fast growth of the Internet and social network, a massive amount of data has been generated and move to the cloud. With the ability to separate the control and data plane, SDN (Software Defined Network) provide an emerging solution for data transportation management tasks in the data center. In recent years, more literature focused on using SDN to manage data center network. It has been shown that SDN switch can support fine-grained rule matching with more than 12-tuple flow. However, Ternary Content Addressable Memory (TCAM), which used to store the flow table in SDN switch, has limited capacity and power-hungry. The performance of the data center throughput would reduce dramatically due to flow table overflow. Previous literature proposed two kinds of solutions, rule replacement and rule caching. In this paper, we propose a new rule caching method based on Long short-term memory (LSTM) to improve the cache hit ratio in SDN switches. From the experiment result, we surprisingly find that the deep learning based prefetching model can predict future flow rules with high accuracy. And then improve the cache hit ratio on TCAM compare with the famous FIFO and LRU cache.