Building the Software Defined Network

Sean Varley is the Director of Planning and Strategic Marketing for the Software Defined Networking Division at Intel –

The software-defined networking (SDN) movement has been through many permutations over the last couple of years, but the latest migration toward HW and SW separation may make the most sense of all.  The fully integrated switch is a stalwart workhorse, serving the industry well for many years, but changing use cases and innovation at “cloud speed” have brought about a requirement for every piece of data center equipment to be flexible, programmable and provisioned.  Today’s announcement from the Open Compute Project detailing bare metal switch contributions from the networking industry brings networking closer to the provisioned model that has fundamentally changed the compute sector and brought about Cloud in the first place.

Today Intel has continued in our commitment to providing partners with open networking components by contributing a generic rack mount 10/40 GB switch specification that is based on our Intel® Open Network Platform Switch Reference Design.  Intel and our partners have been supplying these types of bare metal switches for most of 2013. Now, with the contribution to the OCP working group, we are renewing our effort to expand the open network genre further.

The bare metal switch is one step in a progression that could show the way to a “server-like” model where choice is embraced at each layer of the system.   End users have been searching for more choice in their network infrastructure, but tight integration between HW and SW have been a barrier to adoption of the server OS model, where more common products and tools from a variety of vendors can come together on off-the-shelf HW solutions.  To be fair, the tight integration of the network system exists for a reason: networking equipment is designed first for performance, no matter what functionality is being implemented.  In order for the bare metal switch to thrive, general purpose operating systems with high performance control planes based on standard, server OS programming models must be present.

Fortunately, this ecosystem is evolving rapidly; several entrants based on Linux* are now coming to market and are embracing Intel® Architecture strengths in virtualization, IO, memory and more.  Intel has led in this area since early this year by offering a networking OS based on Linux with abstractions and APIs meant to enable an extensible control plane build.   Loadable network OS’s combined with network boot options, such as the open source ONIE project, are fundamentals in the evolution of a networking market that offers more choice to end users.

Barriers to the adoption of bare metal switching don’t stop at the networking OS; the control plane must also be flexible and portable.  The dearth of standard interfaces for network control planes will limit the portability and interoperability required to speed bare metal switch adoption.  Not surprisingly, this is a burgeoning area of development in the industry.  Beginning with the thread started by standards such as OpenFlow, efforts are now moving forward to programmatic layers that allow for the entire switch to be modeled and configured by rapidly evolving cloud control and orchestration layers utilizing REST or REST-like interfaces.  While the Intel® Open Network Platform APIs are examples of such interfaces, more work is required here to supply a truly interoperable and complete solution.

Intel has been committed to choice in all layers.  To that end, Intel has provided and will continue to provide tools and capabilities in the Intel Open Network Platform family of reference designs, allowing both SW and HW choice in the network ecosystem.  Intel is committed to expanding the available options to a market looking for flexibility, portability, lower cost and ultimately the kind of provisioned solution that will drive the innovation required to keep pace in a rapidly changing SDN and NFV market shaped by the Cloud.etworkion