I found this SDN company at the recent redhat conference
I found this Virtual Network Infrastructure company at Redhat last week. plumGRID. It was interesting because they can demostrate virtual networks on physical, kvm, vmware - the idea was an IO engine gets installed and anywhere the IO engine is it can carry the virtual networks. The fabric can be anything, L2, L3, even IB, and tunnels are in VXLAN but could be xGRE or STT - doesnt matter what the physical is. This way you can do L2 storage across a L3 fabric, and you can keep L2 adjacency off of L3 fabric, and you dont need to deal with any multicast stuff either for learning it. The nice thing here is they have a compiler which allows the compilation of virtual networking entities in the virtual networking domain(s) - so the networking functions that they provide could be changes if needed. The other thing I like here is the vxlan IDs are totally managed. No more VLAN or VXLAN mapping, just create a fabric, a management, and have some external networks and plug them in. Then they can slice up networking the same way we slice up big machines into VMs. You can use the fabric to make as many networking domains as you want and the tenents, whether they are in vmware of openstack, are able to see the networking they are assigned and whatever external legacy networking entities they were given. They also run their IO on top of non-virtual, namespaces, containers - not just VMs. The nice thing about all this complexity is that the physical switches never get touched or played with and it gives me a real opportunity to lower the netops tickets required to make tons of networks - you dont need to netops to do things to make networking slices - I keep thinking slicing like deli meat. The think that worried me is all this complexity can be hard to deal with having networks on networks but the workflows in this product are very interesting as they can be completely hidden behind opestack networking and completely integrated with vmware workflows. I wish Cisco would buy these guys or something as this is what cisco's virtual networking strategy should be but I guess it doesnt help sell hardware so we are going to get asked to rip out all our 6500s to get datacenter features yet again. Seems to me with companies like this talking about virtualizing vi hyperv, namespaces, containers, kvm, openstack and ESXi all with the same IO engine at a Redhat conference and working with redhat openstack to do all this is very interesting to me. I wish I could download a free trial of this stuff but ill be getting a copy of this soon hopefully - have to beg the CIO for time for this. they also have this feature that per virtual networking area per vswitch per router per nat you can have analytics queries pumping the virtual network interface data out, its like having splunk+snmp+netflow+sflow and they support triggers you could build heuristics off this stuff.