Related Documentation
Understanding Reflective Relay for Use with VEPA Technology
Virtual Ethernet Port Aggregator (VEPA) technology aggregates packets generated by virtual machines located on the same server and relays them to a physical switch. The physical switch then provides connectivity between the virtual machines located on the server, so the virtual machines do not communicate with one another. Offloading switching activities from a virtual switch to a physical switch reduces the computing overhead on the virtual servers and takes advantage of the security, filtering, and management features of the physical switch. Reflective relay, also known as “hairpin turn,” enables the physical switch to receive aggregated packets from the virtual machines hosted on the server through the VEPA on the downstream port and send those packets out the same downstream port from which the physical switch received them.
VEPA
Even though virtual machines are capable of sending packets directly to one another, it is more efficient to pass these aggregated packets from the VEPA to a physical switch. The switch can then send any packets destined for a virtual machine located on the same server to the VEPA.
Reflective Relay
Reflective relay, also known as a “hairpin turn” or “hairpin mode,” returns aggregated packets to the VEPA by using the same downstream port that initially delivered the aggregated packets from the VEPA to the switch. Reflective relay must be configured on the interface located on the physical switch that receives aggregated packets, such as VEPA packets, because some of these packets might need to be sent back to the server if they are destined for another virtual machine on the same server.
Reflective relay only occurs in two situations:
- When the destination address of the packet was learned on that downstream port
- When the destination has not yet been learned
Reflective relay does not otherwise change the operation of the switch. If the interface to which the virtual machine is connected and the MAC address of the virtual machine packet are not yet included in the Ethernet switching table for the virtual machine’s associated VLAN, an entry is added. If the source MAC address of an incoming packet under the respective VLAN is not yet present in the Ethernet switching table, the switch floods the packet on all the other ports that are members of the same VLAN, including the port on which the packet arrived.
