Calculate VXLAN encapsulation overhead and VNI address space.
VXLAN wraps each original (inner) Ethernet frame inside a new outer packet: an 8-byte VXLAN header identifying the virtual network, an 8-byte UDP header, a 20-byte outer IPv4 header, and a 14-byte outer Ethernet header — 50 bytes of overhead in total. Because the encapsulated frame must still fit within the physical network's MTU, the effective usable MTU for guest/inner traffic is the physical MTU minus this overhead, which is why VXLAN deployments commonly require jumbo frames (MTU 1600+ or 9000) on the underlay to avoid fragmenting normal 1500-byte guest frames. Separately, VXLAN's 24-bit VNI field supports over 16 million isolated virtual networks, vastly more than the 4096 VLANs traditional 802.1Q tagging allows.
Effective MTU
effective_mtu = physical_mtu − 50
Max VNIs
max_vnis = 2^24 = 16,777,216
With a standard 1500-byte physical MTU, VXLAN's 50-byte overhead leaves only 1450 bytes for the inner frame — below the standard 1500-byte Ethernet MTU guest VMs expect, which forces fragmentation or requires the guest OS to use a smaller MTU. Enabling jumbo frames (e.g. 1600+ or 9000-byte MTU) on the physical underlay avoids this by leaving enough headroom to encapsulate a full-size 1500-byte guest frame without fragmentation.
Traditional 802.1Q VLAN tags use a 12-bit ID field, capping the address space at 4096 VLANs — insufficient for large multi-tenant cloud environments. VXLAN's VNI field is 24 bits, providing over 16.7 million possible network identifiers, which is why VXLAN became the standard overlay technology for cloud-scale network virtualization.
Yes, marginally — the 50 bytes of overhead per packet reduces the payload-to-overhead ratio for small packets more than large ones, and if fragmentation occurs due to insufficient underlay MTU, throughput can degrade significantly more than the raw overhead percentage suggests, since fragmentation/reassembly adds CPU cost and potential packet loss.