In modern EVPN-VXLAN multi-fabric and Data Center Interconnect (DCI) architectures, encapsulating Ethernet frames across routed underlay boundaries introduces protocol overhead that demands systematic Maximum Transmission Unit (MTU) engineering.
VXLAN encapsulation adds a 50-byte (or 54-byte with dot1q) outer header, consisting of the Outer Ethernet Frame, Outer IP header, UDP header, and the 8-byte VXLAN Network Identifier (VNI) header. Furthermore, multi-fabric architectures frequently traverse Border Gateways (BGWs) or intermediate transport networks that may append additional label stacks or encapsulation headers (such as MPLS, EVPN Type-5 route metadata, or MACsec encryption trailers).
To completely prevent packet fragmentation—which severely degrades ASIC packet processing rates, increases CPU interrupts, and breaks path MTU discovery for overlay endpoints—the physical underlay transport infrastructure must support full jumbo frames. HPE Aruba Networking ' s best-practice standard for multi-fabric data center underlays explicitly specifies configuring a system-wide underlay MTU of 9198 bytes. This ensures that standard 9000-byte tenant jumbo payloads, hypervisor vMotion frames, and storage iSCSI packets pass across multiple encapsulated fabric boundaries without truncation or fragmentation overhead.
Reference Topics: HPE Data Center Network Design – EVPN-VXLAN Underlay Engineering; Jumbo Frame Sizing and 9198 MTU Configuration Standards.
===============
Submit