It’s been some time since I wrote something. Not because I got lazy but because I’ve been cooking something.

We all know and love OpenStack and recently they decided to deprecate the Linux-Bridge Neutron ML2 Driver and replacing it with OVN to provide a more future proof way of providing overlay networking and dynamic networking.

OVN, or Open Virtual Network, is practically just OpenVSwitch with some management and clustering tooling.

There used to be a bug in OVS which made it impossible to use for setups with many routes as OVS subscribes to route updates of all routing tables.

If you have full bgp tables this easily ends up with 937k routes per upstream if you utilize multiple VRFs.

Luckily Ilya had some spare time at his hands and worked with me on a suitable solution for this and we were able to finally close #185 !

Now instead of crawling through all routing tables it will only care about the default routing table, making all other routing tables “inaccessible” to OVS for routing/switching decisions.

This has little to no drawbacks apart from very rare setups which utilize multiple routing tables but expect packets to be switched between those tables. This sort of contradicts Switching/Routing however.

Ok that’s all fine and cool but what’s the big deal here?

I dont blame you for skipping the whole intro part and going straight to here. I’d do the same really.

The big deal is now that you are able to have fully collapsed Networking and Compute nodes directly at the edge.

You can now place a full BGP member in your datacenter armed with high density Epycs or Xeons and provide virtual machines that reside directly at the demarcation point.

Single Hop to the internet without going through a router that could congest or introduce latency.

The effect for me is quite notable, I use comodity hardware and all of my VMs have now linerate 10Gbit/s connectivity despite the fact that I’m using a Software Router and Off-Brand NICs/DACs/Chassis.

In fact I can even go 40Gbit/s if I would bother buying better NICs!

And yes all this is without SR-IOV or DPDK which would obviously make this fly even faster!