Design of TCP Congestion Control in Data Center Networks Based on Stable Round Trip Time

Abstract

Data center networks have become increasingly important over the years as they are what make cloud computing, mobile computing, machine learning, and artificial intelligence possible. TCP is the main transport layer protocol in the Internet. Its congestion control has evolved for decades, with many versions or variants having been proposed and improved. Among them, several well-known TCP versions have been implemented either in different versions of the OS kernel or in the data center. Whether it is based on packet loss, round-trip-time (RTT), or bandwidth-delay product (BDP), each has its own advantages and disadvantages. In this paper, we will review the characteristics of the data center, and then propose a TCP version that is more suitable for the data center and achieves reliable transmission with high throughput and low latency.