The DLR is optimised for forwarding in the logical space between VMs, on VXLAN-backed or VLAN-backed portgroups.
The DLR has the following properties:
High performance, low overhead first-hop routing:
Scales linearly with the number of hosts
Supports 8-way ECMP on uplink
Up to 1,000 DLR instances per host
Up to 999 logical interfaces (LIFs) on each DLR (8 x uplink + 991 internal) + 1 x management
Up to 10,000 LIFs per host distributed across all DLR instances (not enforced by NSX Manager)
Keep in mind the following caveats:
Cannot connect more than one DLR to any given VLAN or VXLAN.
Cannot run more than one routing protocol on each DLR.
If OSPF is used, cannot run it on more than one DLR uplink.
To route between VXLAN and VLAN, the transport zone must span single DVS.
The DLR’s design at a high level is analogous to a modular router chassis, in the following ways:
ESXi hosts are like line cards:
They have ports with connected end stations (VMs).
This is where the forwarding decisions are made.
The DLR Control VM is like a Route Processor Engine:
It runs dynamic routing protocols to exchange routing information with the rest of the network.
It computes forwarding tables for “line cards” based on the configuration of interfaces, static routes, and dynamic routing information.
It programs these forwarding tables into the “line cards” (via the Controller Cluster, to enable scale and resiliency).
The physical network connecting ESXi hosts together is like a backplane:
It carries VLAN-encapsulated or VXLAN-encapsulated data between the “line cards.”