OpenVSwitch has become the standard for software/flow switch implementations in the industry and performs a variety of hardware offloads to proprietary vendor components transparently to the consumer.
Adapting VyOS to
- enable OVS bridges (if the user selects to do so - purely optional until proven-out)
- allow them to offload
- making the control plane aware of this when
- binding VXLANs or other "acceleratable" datapaths
- deploying rules/filters which can be offloaded to HW
- functions which can be handed to a hardware switch (such as those in ConnectX cards)
- allowing use of or implementing a centralized method for OpenFlow control & telemetry
will help significantly when dealing with high-performance environments handling 100/400G and future fabrics.
Future integrations of OVS with various dataplane technologies will therefore allow VyOS to piggyback on those advances yielding new functionality and performance quotients as a byproduct "for free."
Flow switching by itself is a very useful function and having a centrally-controlled flow fabric of VyOS devices in physical and virtual space would be extremely handy. Throw in the topology properties of OVN and it becomes feasible to run an entire cloud (EW, NS, tenant, and service fabrics) entirely on VyOS handling a variety of (offloaded) hardware and virtual environments (such as the edge "devices" used in cloudstack/opennebula).