• United States

3 strategies to simplify complex networks

Sep 16, 20194 mins
Network Management SoftwareNetworkingSD-WAN

Innovations such as SD-WAN, Wi-Fi 6 and 5G have enabled networks to do more, but they’ve also made them complex. Software, machine learning, and automation will alleviate the problem.

A network of digital connections extend broadly across a city.
Credit: Metamorworks / Getty Images

As the cloud era meets the demands of digital transformation, networks must change. That means for enterprises, they must become simpler, said Juniper CEO Rami Rahim, speaking at the company’s annual industry analyst conference last week.

The past five years has seen more innovation in networking than in the previous 30. Things such as SD-WAN, multi-cloud, Wi-Fi 6, 5G, 400 Gig, and edge computing are on the near-term horizon for almost every company. While all of those technologies have enabled the network to do so much more than ever before, their complexity has also risen.

juniper ceo rami rahim Zeus Kerravala

Juniper CEO Rami Rahim

Network engineers face the harsh reality that they are being tasked with working faster but also more accurately to cut down on unplanned downtime. Networks must become simpler to run, which actually requires more engineering from the vendor. Think of the iPhone. It’s so simple, my dad can use it without calling me every hour. Making it easy requires a tremendous amount of innovation from Apple to mask the underlying complexity.

How to simplify networks

Vendors can help make networks simpler by executing on the following:

  • Simplicity through software. The pendulum has swung way too far on the “hardware doesn’t matter” theory. Of course it matters, particularly for networking where tasks such as deep-packet inspection, routing, and other functions are still best done in hardware. However, control and management of the hardware should be done in software because it can act as an abstraction layer for the underlying features in the actual boxes. For Juniper, Contrail Cloud and their software-delivered SD-WAN provides the centralized software overlay for simplified operations.
  • Machine learning-based operations. Networks generate massive amounts of data that can use useful for operating the environment. The problem is that people can’t analyze the data fast enough to understand what it means – but machines can. This is where network professionals must be willing to cede some control to the computers. The purpose of machine learning isn’t to replace people, but to be a tool to let them work smarter and faster. Juniper acquired Mist Systems earlier this year to provide machine learning based operations to Wi-Fi, which is a great starting point because Wi-Fi troubleshooting is very difficult. Over time, I expect Mist’s benefit to be brought to the entire enterprise portfolio.
  • Vision of intent-based operations with purposeful automation. The long-term goal of network operations is akin to a self-driving car where the network runs and secures itself. However, like with a self-driving car, the technology isn’t quite there yet. In the auto industry, there are many automation features, such as parallel park assist and lane change alerts that make drivers better. Similarly, network engineers can benefit by automating many of the mundane tasks associated with running a network, such as firmware upgrades, OS patching, and other things that need to be done but offer no strategic benefits.

To ASIC or not to ASIC

As I mentioned, network hardware is still important. There’s currently a debate in the network industry as to whether companies like Juniper should be spinning their own silicon or leveraging merchant silicon. I believe ASICs allow vendors to bring new features to market faster than waiting for the silicon vendors to bake them into their chips. ASICs also give the network equipment manufacturer better control over product roadmaps.

However, there is a wide range of silicon vendors that offer chips for a multitude of use cases that might be hard to replicate in custom chips. Also, some of the cloud providers know the specific feature set they are looking for and will dictate they want something like a Barefoot – Tofino-based switch. In this case, merchant silicon would provide a time to market advantage over custom. But both approaches are viable as long as the vendor has a clear roadmap and strategy of how to take advantage of hardware and software.

Historically, Juniper has done a great job using custom chips for competitive advantage, and I don’t see that changing. Buyers should not shy away from one approach or the other. Rather they should look at vendor roadmaps and choose the one that meets its needs best.

There’s no shortage of innovation in networking today, but new features and functions without simplicity can wreak havoc on a network and make things worse. One of my general rules of thumb for IT projects is that the solutions must be simpler than the original issue, and that’s not often the case in networking. Simplifying the network through software, machine learning, and automation enables businesses to take advantage of the new features without the risk associated with complexity.


Zeus Kerravala is the founder and principal analyst with ZK Research, and provides a mix of tactical advice to help his clients in the current business climate and long-term strategic advice. Kerravala provides research and advice to end-user IT and network managers, vendors of IT hardware, software and services and the financial community looking to invest in the companies that he covers.

More from this author