LIGHT-LB: A Lightweight Load Balancing Strategy for Small Flows in Data Center Networks

With the rapid growth of cloud computing and large-scale distributed applications, data center networks (DCNs) are facing increasing demands for high throughput and low latency. Traditional Equal-Cost Multi-Path (ECMP) routing lacks real-time congestion awareness, resulting in suboptimal performance for small flows. In response, solutions like CONGA and HULA have emerged, but they introduce deployment complexity and additional network overhead. This paper proposes LIGHT-LB, a lightweight load balancing mechanism implemented using P4 programmable data planes, based on per-hop latency feedback and reverse path learning. LIGHT-LB minimizes state maintenance and probing costs while achieving efficient deployment. Experimental results show that LIGHT-LB significantly improves small flow completion time (FCT) under various traffic loads, performing comparably or better than HULA, while maintaining scalability and deployment feasibility.