A Novel Least Connection-First Stateful Load Balancer for TCP Traffic

This paper presents the Least Connection-First algorithm, commonly employed in TCP, as a load balancing mechanism implemented within a P4 switch. While previous work in SilkRoad demonstrated the feasibility of P4 switches for cost-effective, high-speed Layer-4 stateful load balancing using hash-based algorithms, these approaches assumed uniform server workloads, making them susceptible to hash collisions and potential connection initiation failures. In contrast, our research leverages the inherent characteristics of TCP connections – specifically, the correlation between request frequency and connection duration – to enhance load balancing performance. By leveraging the P4 switch’s ability to identify SYN and FIN packets in the data plane, our approach enables the forwarding of these control packets to a central controller. The controller then dynamically assigns incoming connections to servers with the fewest active connections. We conduct our experiment on a P4 hardware switch for performance validation. Experimental results demonstrate that our proposed method exhibits superior performance compared to traditional hash-based and query-based Least Connection-First schemes.