By Ben Grubin
HP’s announcement last week at Interop that they are shipping their SDN SDK and SDN App Store is merely one of the first salvos of a war that will likely heat up over the next 24 months. What was once the purview of marketing and start-ups, Software Defined Networking has now become the dominant strategy of HP, VMware and others to truly disrupt the current state of data center network architecture.
As they announced a few months ago, VMware is going further with their NSX-based SDDC (Software Designed Data Center) concept, which essentially treats the entire underlying network infrastructure as dumb pipes.
In this new world of SDN and SDDC, the never-ending list of features that Cisco, HP, Dell, Huawei, and others have used as the lynchpin of their competitive strategy in the Ethernet switching and routing markets is nearly irrelevant. Instead, what these new technologies demand is simplicity and speed, something incompatible with layering on hundreds of unnecessary features into the software that drives Ethernet switches.
In fact, layering is the underlying story here. While most network architects have tried to avoid networking overlays due to complexity and losing visibility into layer 2 and 3 architecture, SDN and SDDC are truly a network overlay that abstracts away the entirety of the underlying physical network.
Implementing SDDC means only two basic requirements for the underlying network: it should have as few hops as possible between any two points, and as much “symmetric” capacity as possible–meaning the capacity should be equally large between any two points on the network. Only with this design do you enable the broadest possible freedom at the overlay SDDC layer.
What don’t you need? VLANs, layer 3 routing protocols (OSPF, IGRP), and other such mainstays of the data center. All of this is handled inside the software layer, and with VMware NSX the virtual infrastructure.
All in all, this is an exciting movement towards simplifying the network layer and making it more agile and responsive to the needs of business. Having per-VM virtualized network components such as load balancers, firewalls, and switches means less specialized equipment and less capital outlay in the racks.
Is all of this going to be in production tomorrow? No way. There’s still some key hardware and software challenges that need to be solved to equalize the performance equation. However, if history is our guide, it won’t take long for those to be conquered.
John Treadway 