skvoor - Fotolia

Evaluate Weigh the pros and cons of technologies, products and projects you are considering.

Physical network infrastructure to decrease in SDN virtual evolution

Ethernet switches, Wi-Fi APs and routers may be the only physical network infrastructure you need in a virtual-, SDN- and cloud-based future, according to networking expert Craig Mathias.

Virtualization is usually discussed in the context of virtual machines as a way of dividing up the enormous power of contemporary microprocessors among simultaneous but unrelated operating environments. But virtualization can be applied more loosely to mean substituting services, even those provisioned in the cloud, for capabilities historically deployed on local servers or other physical network infrastructure.

So, the question arises: Just how much physical network infrastructure could be virtualized in this manner both for enterprises and network operators?

The answer in both cases is: almost everything. In fact, the term network functions virtualization describes exactly what we have in mind here. Obviously, a few issues need to be considered before virtual network infrastructure becomes common, but the implications are dramatic regardless.

A network in the cloud? Network as a service? Yes to both. I'm betting this is the path forward for almost every network everywhere, with software-defined networking providing the incentive to innovate and bring about many of the resulting benefits.

This brings us to a question I've been asking clients and others lately: What does the network of 2020 look like? Or for the more adventurous, what will the network of 2025 look like?

To get an idea, let's look at the key trends that are gaining traction:

The savings on local equipment and physical plant might easily pay for a strategy that is preferable regardless.

Looking at these influences, it's easy to see precisely how physical network infrastructure will evolve and consolidate over the next 10 years. For example:

  • Wi-Fi access points will remain the primary and default access for almost every client device, from handsets and tablets to the Internet of Things. Every available bit of radio spectrum at 900 MHz and 2.4, 5 and 60 GHz will be used, with the resulting capacity sufficient for almost every application. We'll even see 10 GbE over the air.
  • Consequently, Ethernet switches will quickly, cost-effectively and, perhaps surprisingly for some, evolve to 10 GbE as the default, with 1 GbE no longer adequate, and interest in 2.5 and 5 GbE rapidly diminishing.
  • Router functionality will be built in as an option in those very same Ethernet switches, much as we see in residential-class routers today. And, yes, resilience and fault-tolerance across the board will become standard and relatively simple.

Here's the surprising part: That's it. That's all the "real" infrastructure required.

What happens to the rest of the physical network infrastructure?

Everything else -- network management, performance optimization, analytics, security and the SDN controller itself -- will be virtualized and provisioned in the cloud. That cloud might be public or private, but public cloud will increasingly be the path of choice given the cost-effectiveness, reliability and on-demand scalability inherent in the public cloud model. Given very-high-performance links to the Internet (10 Gbps or more will be common), there's no performance penalty compared to local implementations. The savings on local equipment and physical network infrastructure might easily pay for a strategy that is preferable regardless.

As continues to be the case, SDN is both the catalyst and the driver for this vision becoming reality. The essence of SDN is in the fundamental programmability of network infrastructure, and the less diversity in this infrastructure the better.

Scalability means adding more of the same elements. Again, these elements are just APs and switches -- APs and switches based on SDN, that is. The inherent performance of advanced processors and interfaces coupled with the flexibility and extensibility inherent in SDN then appear to be key guideposts on the path to networking nirvana.

Our work to date here is hardly exhaustive. And change, even in technology, occurs more slowly than most of us would prefer. But there's little denying that moving toward SDN is perhaps the most important direction overall in networking today. Coupled with the evolution of hardware capabilities discussed above, it's easy to see how the expanding role of SDN is bringing along a broad range of additional benefits.

Next Steps

Network hardware's role in virtualization and overlays

Network infrastructure will evolve, not face its demise

Managing physical and virtual networks in parallel

This was last published in May 2016

Dig Deeper on Network hardware and SDN



Find more PRO+ content and other member only offers, here.

Join the conversation


Send me notifications when other members comment.

By submitting you agree to receive email from TechTarget and its partners. If you reside outside of the United States, you consent to having your personal data transferred to and processed in the United States. Privacy

Please create a username to comment.

What do you think your emerging virtual and physical network infrastructure will look like in 2020?
Actually, this is a pipe dream. It's like the astronomer's and astro-physicist's futile search for the "edge ofvthe universe" and other "earth-like planets" that "may contain life".

Here's the problem: Moore's law and Quantum Computing. Moore's law will still be relevant for several years, and when we finally reach the saturation point where it will no longer be possible to double the ASIC speeds while reducing the ASIC footprint, every 2 years, then Quantum chipsets will be available, and we'll pick up where we left off.

Wha does this mean? It means that hardware forwarding will ALWAYS be much faster than software forwarding of packets. Tricks used in the VNF world, like use of high speed SmartNICs with SR-IOV capability, or DPDK at least, are just attempts to create a hybrid where forwarding continues as close to the hardware as possible.

So, go ahead and orchestrate your VNFs inside of virtual machines or containers, and be as elastic as you want to be, but, in the end, because enterprise customers like broadcasters want RT CoS, and cannot tolerate packet loss, higher latency, and higher jitter. You are going to learn the hard way that there is a price to be paid for elasticity.

That's why in the end most will converge to a hybrid, where the data plane will ride thru ASICs, and, depending onnthe convergence requirements of the control plane, the controller may or may not reside totally in software.

Time to smell the coffee and crawl out of our "Virtual Dreamworld".