It’s great to see, hear and read various points of view on the evolution of networking. It’s a hot topic right now, highlighting the fact that the network is at the center of the market transitions driven by Mobile, Cloud, new breeds of Apps and the Internet of Things. Technical leaders from my team have become road warriors recently, talking to customers, media and investors about the evolution in networking, sometimes referred to as Software Defined Networking (SDN)
There’s a healthy debate in the market about SDN, and with any debate comes confusion. SDN’s initial definition (the logical separation of routing and switching control plane and data plane) has been stretched so far that it has come to mean something different to everyone.
There are plenty of use cases driving the attention that SDN is receiving today. For instance, Service Providers are looking at trends like Network Functions Virtualization for network elasticity as an opportunity to create greater business value by launching new services quickly. Traditional enterprises think about SDN as a way to rein in the operational and management complexity of data centers to scale infrastructure. Academic institutions want open source controllers, so they can economically slice campus networks for both production and research purposes. At least one thing is crystal clear: one size does not fit all when it comes to deploying SDN.
In some circles SDN has become synonymous with the erosion of value in the underlying networking infrastructure – the hardware and the ASICs. There is an argument purporting that when network intelligence is abstracted into software, hardware and silicon innovation will become less important and even commoditized.
I’m going to take this opportunity to address these misperceptions about the changes taking place in networking with three truths about the next chapter in networking as Cisco sees it.
1. Network Virtualization is Different than Server Virtualization
Underutilized compute resources created the opportunity for server virtualization. Companies were commonly running a single operating system image and a single application on a high performance server, and as a result they were only using 10-15% of the server’s computational resources. Hypervisors allowed people to host multiple operating systems and applications on the same server, which drove greater utilization and efficiencies by creating a shared resource.
However, underutilization is not a problem in the network – the truth is far from it. The network has always functioned as a shared resource – basic technologies like VPNs allow people to share the same infrastructure. And, in reality, the network is highly utilized and highly efficient. Here is a more detailed analysis of the difference between server and network virtualization by industry analyst Peter Christy.
Put simply, today’s businesses demand more capacity from the network – not more consolidation. In fact, server virtualization is pushing the limits of today’s network utilization and therefore driving demand for higher port counts. Server virtualization is actually creating the demand for higher capacity networks, as well as the opportunity for value engineering at the systems level with next generation switching fabrics.
The forces driving the adoption of network virtualization are very different than forces that drove the adoption for server virtualization. As opposed to focusing on consolidation, today’s businesses want to turn on new services and applications faster, as well make networks easier to provision, manage and re-configure based upon needs of applications at scale. This application-centric approach to networking places increased importance on the entire infrastructure stack working together in order to lower operational hurdles and expenses, while providing the optimal performance of services and applications.
2. Application Performance Creates Greater Demand for Physical and Virtual Innovation
The transition happening in IT and the consumer Internet can be summed up as a shift from a Web-based to an App-based economy. In this transition, we’re seeing the emergence of new breeds of applications, and network performance is becoming more critical. As user expectations and business requirements demand more from the applications, optimal performance can only be achieved through combined innovations in ASICs, hardware and software. Today’s app economy requires the performance driven through hardware innovation, with the scalability that comes from virtualization, coupled with customized ASICs to deliver the best system to run an application. SDN doesn’t obviate the need for value engineering in hardware – rather it magnifies its importance in a systems approach because commodity hardware simply can’t offer the necessary performance. Ultimately, the App-based economy requires a significant shift in the IT model of today.
The ability for systems engineering to deliver better application performance can be seen in a recent example of Cisco-led innovation. As Hyper-V, Red Hat KVM and Xen grow their market share, at the expense of VMware, commoditization is actually happening at the hypervisor. The hypervisor is nothing more than a feature in the cloud stack. Against this backdrop, Cisco’s Unified Computing System (UCS) has grown to be a US$ 2 Billion business in four short years. What’s interesting about this example is that UCS makes use of virtualization; it exploits the hypervisor to deliver a better management experience. While virtualization did commoditize generic server offerings, it also made the market opportunity that Cisco seized by engineering UCS. In other words, the disruption that was server virtualization, created the opportunity to innovate at the systems level as evidenced by UCS.
By the way, Cisco isn’t the only one to see the value of hardware in the Software Defined Networks of tomorrow. Here is a slide that IDC’s Vice President of Network Infrastructure Rohit Mehra presented recently at their annual Directions conference:
To summarize, of the $3.7B SDN market opportunity represented, 58% of the market or roughly $2.1 Billion is network infrastructure. Neither routers nor switches will become irrelevant in the world of SDN. Only 8.7% or roughly $330 Million represents the market opportunity for the control-plane enabling SDN technology – whether that’s a software controller or some form of virtual overlay technology. The truth is, Cisco will see a $670 Million market for next generation applications and network services that will get created on top of the hardware and software infrastructure. By the way, this market is an incremental opportunity to the business Cisco operates in today. Not to mention a high margin incremental business in the roughly $570 Million market for services that tie the new applications to the systems infrastructure.
3. A Software-Only Approach Actually Increases Cost
Now, let’s take a look at the perceived benefits brought about by server virtualization. Again, here’s a popular slide from industry research firm IDC:
This slide shows that since the advent of mainstream x86 server virtualization, the operational costs of the data center have soared to approximately 80% of IT budgets. While virtualization led to a consolidation of commodity servers, it aggravated management costs due to the virtual machine sprawl it created. The truth is that virtualization on top of legacy commodity infrastructure doesn’t pay off at the server level. There is no compelling reason to repeat this mistake and add operational complexity to other parts of the data center. Instead, to contain operational costs what’s needed is the blending of hardware and software innovation akin to what UCS delivers. It bears repeating that while the commodity server market roughly stayed flat during the rise of virtualization, UCS with its value engineering grew to a $2 Billion dollar business with 20,000+ customers in four years thanks to innovation on top of virtualization.
The next chapter of networking will be a systems approach – ASICs + hardware + software – all working together. The entire system must be programmable via APIs where appropriate, and it must integrate into existing and emerging cloud stacks. It requires central policy management and the agility to scale for application delivery to users with end-to-end security. That’s what Cisco is building. Software-alone just doesn’t cut it.
Those are the three truths about networking as we see it. True innovation will only result from the combination of ASICs, hardware and software that use the inherent network intelligence to provision, manage and use applications more efficiently. But that’s another blog for another time…