VXLAN/BGP EVPN for Trading: Multicast Scaling Challenges for Trading Colocations

Abstract

Financial exchanges today depend on ultra-low-latency market-data delivery, so the underlying network architecture is no longer an afterthought but a strategic asset. In colocation centers where nanoseconds count, VXLAN encapsulation paired with a BGP-based EVPN control plane has become the de facto framework for segmenting and scaling Layer 2 traffic. This paper investigates the role of multicast in that architecture, exposing the technical pressure points network engineers must address to protect the integrity of real-time trading streams. Although EVPN decouples forwarding and forwarding-label assignment, head-end replication still burdens CPU queues on high-density 100-GbE line cards. Multicast is indispensable for distributing top-of-book feeds; however, performance metrics deteriorate sharply once the subscriber count exceeds fifty ports. The study benchmarks the standard draft-based HER against PIM/MBGP and MPLS-VPN, quantifying the round-trip delays, forwarding table churn, and control plane convergence times that each protocol injects into the data path. Experimental findings are paired with field data from production firms, yielding pragmatic recommendations for hardware selection, orderly fan-out patterns, and resilient control-plane adjacency. The paper also reviews nascent tools—segment routing over IPv6, P4 programmable pipelines, and publish-subscribe layers like Apache Kafka—that promise to offload or supplant traditional multicast domains as trading volumes continue their logarithmic growth. By distilling these insights into repeatable design outlines, the article supplies quantitative guidance for architects who must balance performance budgets against regulatory uptime mandates.