Virtualization: Hypervisors, Overlays, and Network Functions
Hypervisors and virtual switching
A Type 1 hypervisor (ESXi, KVM) runs directly on hardware; a Type 2 hypervisor (Workstation, VirtualBox) runs on top of a host OS. Guest VMs attach virtual NICs to a virtual switch. A standard vSwitch is per-host; a distributed vSwitch spans hosts and is configured centrally. VLAN tagging modes matter: virtual switch tagging puts the vSwitch on the trunk, virtual guest tagging lets the guest tag its own frames, and external switch tagging passes already-tagged frames through.
Isolation and overlays
VRF-lite creates separate routing tables on one device without MPLS. GRE gives generic tunneling, IPsec adds confidentiality, and VXLAN provides Layer 2 over Layer 3 with a 24-bit VNI, far exceeding the 4096 VLAN limit of 802.1Q. VXLAN needs a control plane; options are flood-and-learn or an EVPN-based control plane, while SD-Access uses LISP for endpoint mapping.
NFV and cloud
Cisco's virtual platforms include the Catalyst 8000V (formerly CSR 1000V), the ASAv, and Enterprise NFVIS. Cloud vocabulary matters: IaaS, PaaS, and SaaS for service models, public/private/hybrid for deployment models.
Traps
Do not confuse a VRF (routing table isolation) with a VLAN (broadcast domain). VXLAN without a control plane relies on flooding. Anycast gateway lets every leaf own the same gateway IP and MAC, avoiding stretched L2 hairpins. Also remember that vPC and VSS are device-level redundancy, not virtualization.
Worked example: MTU math
VXLAN overhead is 50 bytes (20 outer IP + 8 UDP + 8 VXLAN + 14 inner Ethernet). For a 1500-byte payload the underlay MTU must be at least 1550. Add IPsec overhead and the requirement grows further, which is why underlay jumbo frames are common.