Reimagining The Network For A Hyper-Converged Infrastructure

Editor’s note: Kittu Kolluri is a general partner at NEA, focusing on IT and energy technology investments. His investments include Box, Aerohive Networks, Braintree, BloomReach, VeloCloud and Cohere Technologies, among others. 

You’re only as strong as your weakest link, and that’s the Internet’s biggest problem today. For more than two decades, the very backbone of the Internet — the network — has subsisted in the shadows, pinned beneath the weight of a fast-growing Internet economy and a powerful cohort of incumbents with much to lose.

Since the early days of SaaS, the tech world has set an astonishing pace for innovation. If the cloud breaks everything (and it does), we are nothing if not resilient — toppling each obstacle in a race to deliver anything-as-a-service, big data and mobility. We have evolved into device-addicted, app-devouring gluttons for bandwidth — all the while coaxing and cajoling the underlying architecture to manage a far greater load than it was ever designed to bear.

The impact of this top-down innovation is often illustrated from a wide area network (WAN) perspective — but if the sprawling WAN is a rising tide, the data center is scrambling to shore up against the floodwaters.

We weren’t turning a blind eye; quite the opposite, in fact. But from a technological perspective, the architecture couldn’t simply be fixed, it had to be reimagined. This is no small feat in a complex ecosystem of industry giants, legacy systems, and intra-organizational dynamics. Yet one thing has always been clear: The problem would only be solved by pushing technology forward.

And push forward we did, albeit in silos: compute, storage, and network. Before cloud computing, applications segmented resources into those three tidy groups. This worked great until it didn’t:  With virtualization came the horizontal scale-out of the compute layer.

Now storage had to move closer to compute in order to ensure performance and scalability. Converged storage addressed the proximity issue, as demonstrated by companies like Nutanix, SimpliVity and Springpath but the resulting increase in east-west traffic shifts the bottleneck to the network.

Software-defined networking (SDN) in the form of host-based virtualization (as pioneered by companies like Nicira/VMWare and PLUMgrid), decouples the control plane from the data plane but is not addressing the underlying physical network infrastructure.

In today’s mobile-first, application-rich, always-on world, our approach to infrastructure must be more holistic, and that means a shift to cloud-based architecture. This has long been a dominant theme in NEA’s enterprise portfolio, but over the last year or two we’ve seen the transformation accelerating.

Virtual overlays have rapidly emerged as the favored approach in this nascent market, with some very interesting startup plays and some big moves by incumbents. Despite being tethered to Cisco gear, Cisco’s Application Centric Infrastructure (ACI) stack technology is conceptually similar to virtual network overlays. Interestingly, Cisco now plans to support open source protocols on its switches to make it easier to implement virtual overlays for customers not buying into its ACI stack.

Among startups blazing trails in network OS virtualization, Pluribus Networks stands out for several reasons. Its software-defined fabric allows organizations to converge compute, network, storage and virtualization.

Where earlier solutions relegated the network to a transport layer, Pluribus has built its value proposition around a more strategic, intelligent role for the network. There has been plenty of industry activity validating Pluribus’ open software-defined fabric approach; especially notable is Facebook Wedge.

With both hardware and software options for deployment, the company has raised nearly $100 million to date from an investor group that includes Valley insiders but also strategic global investors. Other startups have begun to pursue a similar model — Alcatel-Lucent spinout Nuage, which started off in the service provider space, focused on network functions virtualization but is increasingly focused on SDN in enterprise data centers. Similarly, Big Switch Networks shifted from host-based network virtualization to a fabric-based model.

Thinking beyond the data center, a number of startups are addressing the widening gap between the enterprise and its branches — some under the umbrella of SDN solutions (i.e. Nuage’s Virtualized Network Services) and others with a direct focus on virtualizing the WAN environment.

Velocloud is among the latter, building the cloud router to solve the problems that point-to-point WAN optimization no longer addresses, and enabling business-grade performance over a public Internet connection.

It seems a bit ironic that as the app world unbundles and functions decouple, the underlying infrastructure is transforming into a software-defined, single-tier mesh fabric — a hyper-converged infrastructure of compute, network and storage. But in light of the explosive growth in traffic and consumption driven by app proliferation, big data, content streaming and online collaboration, it makes sense that the underlying architecture would become increasingly fluid and dynamic.

The shift has been a long time coming, but the solutions are proving simpler and more elegant than anyone expected. To me, it looks like the unglamorous underbelly of the Internet may be the biggest opportunity yet in the cloud transformation.

Disclosure: NEA is an investor in Pluribus, Springpath, Velocloud.