- Few enterprises are 100% virtualized, so trying to make the argument that “a virtual overlay is all an enterprise needs” ignores a very real fact of life.
- Enterprises require intelligence in both the virtual and physical networks, as each plays its part in delivering applications where they are needed.
I had hoped we were done with the smart network/dumb network argument, but I guess nothing ever really goes out of style. History has proven over and over that anyone who tries to set extremes like smart network vs. dumb network is basing their entire premise on unrealistic expectations. If I look at the extremes of either a completely virtual data center – and I mean everything – or a completely physical one, then I can make a convincing argument for either a dumb network or a smart one. However, those examples are the far-edge cases and extremely rare. Enterprises do not exist as edge cases. Enterprises need intelligence in both the virtual network and the physical one.
There are a number of benefits to an overlay network. It offers tight integration with the hypervisor and management systems and means that the network is included in all facets of the application and VM lifecycle. It also enables functions such as service insertion/chaining and injects application services such as load balancing, firewalling, application performance, or any other type of traffic engineering technology directly into the network path.
Of course, virtual network traffic very often has to traverse the physical network, and expecting the physical network simply to forward stuff from here to there is foolish. Enterprise data centers have a bunch of other traffic contending for service. Sure, if an enterprise had unlimited funds, it could easily architect a non-blocking physical network which could be devoid of congestion, but that’s not realistic because the network is not core to driving enterprise business; it is the conduit through which the enterprise business flows. There is traffic between application tiers, traffic between applications, traffic between virtual and physical and virtual servers, traffic between physical servers which is completely opaque to the virtual layer, and traffic between clients and applications. There are conflicting demands on the network and there is going to be congestion.
SDN messaging that downplays the significance of the physical network misses the reality enterprise network architects face. Enterprise network architects have to support both physical and virtual servers and I know of no enterprise that is 100% virtualized. As such, they have to make sure that all the network traffic is meeting various SLAs and expectations from application owners and users, and they cannot do that in a hypervisor. Moreover, in an N tier network, the virtual overlay has no idea what is occurring in the physical network and all it sees is congestion and packet loss.
The fact is, in a virtualized data center, enterprises need an intelligent physical network which provides the forwarding control, congestion management, interconnectivity, telemetry, and connectivity over an application’s lifecycle as well as programmability to support the dynamic needs of competing applications. If the enterprise is using an overlay network, it too needs to be intelligent to maintain connectivity between nodes throughout the application lifecycle and integrate with the physical network for improved traffic engineering and visibility.
On second thought, since a programmable physical network does everything an overlay does, why do enterprises need overlays again?