High-Availability Load Balancing for Bare-Metal Kubernetes Clusters: A Design and Evaluation Study of OpenShift MetalLB with BGP and BFD

Abstract

Bare-metal deployments of Kubernetes-based container platforms such as Red Hat OpenShift lack the native, cloud-provider-integrated load-balancing services that are taken for granted in public cloud environments. MetalLB fills this gap by exposing Kubernetes Services of type Load Balancer on bare-metal or on-premises infrastructure, using either Layer 2 (ARP/NDP) or Border Gateway Protocol (BGP) advertisement modes. This paper presents a design and evaluation study of the BGP advertisement mode of MetalLB as deployed on OpenShift, augmented with Bidirectional Forwarding Detection (BFD) for rapid link and peer failure detection. We describe the underlying protocol mechanics of BGP and BFD, the MetalLB controller/speaker architecture, and the integration points with OpenShift's OVN-Kubernetes network stack. A reference architecture is proposed in which each cluster node peers with top-of-rack switches over eBGP and advertises Virtual IP (VIP) service routes, while BFD sessions running beneath the BGP peering’s shrink failure-detection time from tens of seconds to the sub-second range. A testbed methodology and representative performance results are presented, comparing convergence and failover behavior with and without BFD. The study concludes that combining MetalLB, BGP, and BFD provides a practical, vendor-neutral, and highly available load-balancing fabric for on-premises Kubernetes and OpenShift clusters, at the cost of additional operational complexity in router and cluster network configuration.