Cumulus Linux supports these forms of network virtualization:
- Integrating with VMware NSX
- Integrating with VMware NSX-V
- Integrating Hardware VTEPs with Midokura MidoNet and OpenStack
- Lightweight Network Virtualization - LNV Overview
- Static MAC Bindings with VXLAN
- Ethernet Virtual Private Network - EVPN
- VXLAN Routing
- VXLAN Scale
- VXLAN Hyperloop
- Static VXLAN Tunnels
- Hybrid Cloud Connectivity with QinQ and VXLANs
VXLAN (Virtual Extensible LAN) is a standard overlay protocol that abstracts logical virtual networks from the physical network underneath. You can deploy simple and scalable layer 3 Clos architectures while extending layer 2 segments over that layer 3 network.
VXLAN uses a VLAN-like encapsulation technique to encapsulate MAC-based layer 2 Ethernet frames within layer 3 UDP packets. Each virtual network is a VXLAN logical L2 segment. VXLAN scales to 16 million segments – a 24-bit VXLAN network identifier (VNI ID) in the VXLAN header – for multi-tenancy.
Hosts on a given virtual network are joined together through an overlay protocol that initiates and terminates tunnels at the edge of the multi-tenant network, typically the hypervisor vSwitch or top of rack. These edge points are the VXLAN tunnel end points (VTEP).
Cumulus Linux can initiate and terminate VTEPs in hardware and supports wire-rate VXLAN. VXLAN provides an efficient hashing scheme across IP fabric during the encapsulation process; the source UDP port is unique, with the hash based on L2-L4 information from the original frame. The UDP destination port is the standard port 4789.
VXLAN is supported only on switches in the Cumulus Linux HCL using the Broadcom Tomahawk, Trident II+ and Trident II chipsets as well as the Mellanox Spectrum chipset.
VXLAN encapsulation over layer 3 subinterfaces (for example, swp3.111) is not supported. Therefore, VXLAN uplinks should be only configured as layer 3 interfaces without any subinterfaces (for example, swp3).
Furthermore the VXLAN tunnel endpoints cannot share a common subnet; there must be at least one layer 3 hop between the VXLAN source and destination.
Cut-through Mode and Store & Forward Switching
Cut-through mode is not supported for VXLANs in Cumulus Linux on switches using Broadcom Tomahawk, Trident II+ and Trident II ASICs. Store and forward switching is supported on these ASICs however.
Cut-through mode is supported for VXLANs in Cumulus Linux on switches using Mellanox Spectrum ASICs. However, store and forward switching is not supported on Spectrum.
MTU Size for Virtual Network Interfaces
The maximum transmission unit (MTU) size for a virtual network interface should by 50 bytes smaller than the MTU for the physical interfaces on the switch. For more information on setting MTU, read Layer 1 and Switch Port Attributes.