How will we tame networking complexity?
When I’m speaking with IT leaders, one of the top challenges they are dealing with is rampant complexity in their networks. And it’s no wonder because the demands placed on enterprise networks today are exponentially more than even a decade ago. Indeed, the Internet has never been so critical for business operations. Driven by the massive deployment of SaaS and a hybrid workforce reality, the Internet has become the new enterprise core.
If that weren’t enough to deal with, cyber-attacks are growing more sophisticated and significant day by day. Side note…at Cisco we block more than 170 million malicious DNS queries every single day. But just think of the amount of capacity planning, deployments, configurations, policies, performance monitoring, integrations, hardware and software updates, patches—and the list goes on—that teams have the burden to manage in order to maintain the network.
At the same time the top-of-mind issues of all network engineers is to optimize the application/user experience, and this is no longer limited to the networking layer.
When we ponder the future, if IT teams are having this much difficulty today—how will they fare helping their organizations respond to disruptions and keep pace with the dynamic needs of the business five years from now?
A simple and scary answer is they won’t succeed. IT teams that are already failing to keep pace today will find themselves even further behind when the complexity curve takes an even sharper, upward turn.
Applying predictive analytics to networking
But there’s another answer to be considered. What if networking teams had more leverage? We’re still in the early years of software defined networking adoption, and many organizations have yet to unlock its full potential. At the same time, in my field, we’re researching how we can apply powerful, predictive analytics to software defined networks and the Internet via advanced networking capabilities.
Of course, many businesses like the retail industry have used predictive analytics for years to predict supply and demand, ensuring they have enough widgets on the shelf to meet peaks in customer demand. Interesting enough, the Internet has been exclusively using a reactive approach: once an issue is detected, the network reacts. Why not learn historical behavior to predict future behavior? What if we can anticipate and forecast a network state, know where there will be a failure in a device, client, or application—and then take corrective action before the failure occurs? The result is that the user experience is vastly improved. In fact, the user may never even be aware that there was an issue because proactive remediation kept their applications at optimal performance.
The right combination of data and experience
In order to bring impactful AI/ML (artificial intelligence and machine learning)-enabled insights to network operations teams, large amounts of data are essential.
We’re fortunate that with the world’s largest networking install base, Cisco has an enormous and rapidly growing volume of data to work with. Moreover, we have designed a number of ML-driven products over the past decades in areas such as security, wireless and the WAN, and this is the unique expertise combined with unmatched data that allows Cisco to provide best of breed ML/AI-driven networking technology.
Over the next months, I’ll dive into this broader notion of how predictive analytics applied to networking and the Internet will drastically improve the user experience.
As I begin this series, let’s begin by examining five things you should know about predictive analytics.
1. Algorithms succeed and fail
Algorithms are mathematical constructs that run across massive amounts of networking data, looking at many variables in your network, applications, and Internet paths. It’s important to understand that predictions may be incorrect, and they may not account for some scenarios. However, and just as important, we also can tweak algorithms over time and fine tune them for even superior performance. Even better algorithms are designed to learn (and tweak) by themselves.
2. Many networks improve all networks
By collecting anonymous telemetry data across thousands of networks, the lessons learned from that data can be applied to individual networks. Every network is unique but predictive analytics lets us find where there are similar issues and events and guide remediation. In some cases, algorithms may strictly focus on a given network, whereas for other use cases the algorithm may be trained across a broad set of anonymous datasets thus leveraging even more data.
3. Proactive measures won’t replace reactive measures
Several reactive remediation procedures are based on fixing issues after the failure occurred. Predictive analytics in networking means that you will be able to take proactive measures to prevent issues before they occur. Also notable is that we do combine both proactive and reactive measures for a more robust approach to improving the health of our networks.
4. The cloud makes predictive analytics scalable
The cloud may sometimes introduce complexity to networking as more and more things are moved to the cloud. However, it is also a major reason why we are able to collect large amounts of telemetry data across a very distributed environment. Data flows in from across our global footprint which means our algorithms can take into account variables that may pertain only to a certain country or region. Obviously, security is always a top concern and Cisco follows the highest standards to be GDPR compliant and maintain privacy.
5. We’ll never be able to predict everything
It is simply impossible to predict everything. Why? Because in order to predict, an algorithm must learn some “early signs”. Such signs may not be available in the telemetry, or they may not be available quick enough. And of course, for some failures such early signs may be hidden in the noise. For example, who could predict that a fiber optic line would be cut by a construction crew? Not only would it be an unpredictable event, but a cost-prohibitive and resource-consuming exercise if one were to attempt it. In this particular example, there are few early signs but only a few milliseconds before the failure. Therefore, we should set our expectations that there are limits to predictive analytics, but at the same time, not let that keep us from extracting the full potential value that we can obtain.
After almost 30 years in this industry, I’ve had the true pleasure of working on the development of numerous advanced technologies (IP, MPLS, network recovery, security, Wireless, IoT, …). But I can say that applying predictive analytics to networking is truly one of the most exciting and groundbreaking opportunities that I’ve been fortunate to participate in over my career. Predictive network analytics is a game changer for our industry and we’re on the cusp of bringing these innovations to our solutions at just the right time when networking teams need it most. Why is it a game changer? Simply because we can move the Internet towards a “Predictive Internet” that will drastically improve the user and application experience, period.
I invite you to follow this series over the next several months to dive deeper into this topic. A question for you—if you could solve one problem in your network with predictive analytics, what would you like to take care of first?
- White paper—Towards a Predictive Internet
- #CiscoChat —Possibilities of Predictive Internet
- Cisco Champions Radio Podcast
- Subscribe to Networking blog
As I understand the “Key” to Predictive Internet is Data. What data/parameters is the bare minimum required to start the journey of Predictive Internet?
Comments are closed.