This is a guest article by Ben Ferguson from Shamrock Consulting Group
Smart applications require smart networks in order to function...smartly. Let’s examine one of the most cutting-edge technologies out there – machine learning – and how the need for reliable, cost-efficient processing power has facilitated the development of software-defined networking.
Additionally, we’ll look at how artificial intelligence is itself becoming embedded in SD-WAN solutions as industry leaders and government-backed researchers look to further refine important large-scale IT networks.
Artificial Intelligence and Machine Learning
Artificial intelligence is rapidly expanding its influence within society. It is omnipresent in our daily interactions with businesses, and it is being increasingly integrated into our homes and social lives as digital home assistants and personalized shopping experiences.
Machine learning is a subset of the much wider field of artificial intelligence. Whereas AI encompasses every human endeavor to make machines behave more like ourselves, machine learning focuses specifically on designing algorithms that help computer programs self-instruct. As Tom Mitchell, computer scientist and former chair of the machine learning department of Carnegie Mellon University puts it, “Machine learning is the study of computer algorithms that allow computer programs to automatically improve through experience.”
A classic consumer application of machine learning is in the recommender systems used by Amazon or Netflix, which leverages datasets based on previous individual behaviors (views, purchases, etc.) to predict what those individual users would enjoy and then presents a curated list of suggestions. Although basic machine learning models are grounded largely in pattern-matching, some of the more sophisticated areas of research (e.g. deep learning and neural networks) are likely to lead to the creation of more sophisticated AI applications.
What is SD-WAN technology
Software-defined wide access networks are IT networks that are responsive to the demands of the software applications running on them. The key advantage of SD-WAN is the separation between the routing plane and the control plane. Whereas a traditional WAN can be thought of as an animal blindly following its instincts, the software-defined WAN can look ahead at network conditions and make proactive adjustments to the network path.
Looking at it another way, SD-WANs provide a means of pooling network capacity between connected data centers (and often cloud providers) using centrally-defined rules to respond efficiently to software bandwidth demands. Unlike traditional wide area networks, SD-WANs are managed centrally with all routing rules visible to any engineer accessing the management system. Any rule changes or updates can be immediately applied and will rapidly propagate across the network. This is much more efficient and reliable than manually applying updates and configuring network hardware at each data center.
SD-WAN deployments are increasingly popular among forward-thinking enterprises. According to Shamrock Consulting Group, industry-leading SD-WAN experts, “Few technologies have grown as rapidly as SD-WAN in recent years, with figures from International Data Corporation (IDC) revealing growth in revenue from around $225m in 2015 to more than $800m in 2017.”
The demands of machine learning and other forms of AI are part of the reason for SD-WAN’s meteoric growth.
Why Machine Learning Needs SD-WAN
Machine learning is already influencing the spread of SD-WAN technology because the demands of software applications such as chatbots are too high for traditional hub-and-spoke WANs. These and similar next-generation applications operate in real-time and require agile networks that can instantly provide processing power where needed. Only SD-WANs can provide both the cost-efficiency and quality of service demanded by the businesses investing in these smart applications.
Moving Towards Zero-Touch Automation
While machine learning applications are dependent on SD-WANs, artificial intelligence technology is also being used within SD-WANs to deliver smarter networks. Rather than relying on humans to set parameters, a machine learning-powered SD-WAN can use its vast data processing capability to make its own adjustments based on higher level rules.
Machine learning models can be applied to many different aspects of SD-WAN performance.
Take fault prediction, for example: By integrating machine learning capabilities, SD-WANs can monitor for warning signs that indicate areas of the network that are likely to break and then notify administrators or even make pre-emptive adjustments to eliminate end-user impact. This is analogous to predictive maintenance within the manufacturing industry, whereby components are continually monitored and replaced just before they wear out, extending the life of business assets and avoiding costly downtime.
Network path selection is another area where machine learning could be useful by learning from congestion avoidance feature sets and anticipating future adverse conditions on a specific link. Traffic could then be directed to a more reliable link when similar conditions are recognized.
Machine learning has even been proven to help SD-WANs recognize VoIP traffic. This could be used to optimize existing identification systems to improve quality of service and detect certain intrusions.
With every step along this road, the system moves closer to complete zero-touch automation.
Two examples of industries where the potential of machine learning and SD-WAN integrations are being actively explored are enterprise network services and energy networks.
Example 1: Enterprise Network Services
Enterprise network services is one area where vendors are actively exploring just how much SD-WANs can improve network performance. In both North America and Europe, vendors are experimenting with network management processes such as fault prediction, intelligent edge networking, and WAN path optimization.
Part of the reason for this is that performance is likely to be the main differentiator when enterprises choose their network providers. All SD-WANs serve to free enterprises from the chains of expensive MPLS circuits by offering fast and cheap networking by default.
The top-ranked SD-WAN providers are looking at reliability, optimization, security, visibility, control, real-time analytics and other areas of performance to make their products stand out in a crowded marketplace. As the workforce becomes increasingly mobile and cloud-focused, network services vendors are all trying to position themselves as market leaders based on their cloud-first architectures and ability to span multiple clouds.
Example 2: Energy Networks
Global power grids are aging and inefficient. As fossil fuels dwindle and new ways of generating energy come online, governments throughout the world are looking for ways to use technology to create a new generation of smart grids.
The vision of the future entails smart meters and other edge technology interacting with a responsive network integrated with the power grid. As the concept of the smart energy grid becomes more defined, researchers are looking for ways to refine the IT networks for optimal performance.
The SLAC National Accelerator Laboratory is a case in point: Scientists here have received a grant from the U.S. Department of Energy to help fund network improvements. One of the lines of exploration focuses on improving resiliency through machine learning combined with software-defined networking, which would involve smart outage prediction.
Research studies have been carried out in this area using historical grid data as learning sets to build models with which to train machine learning algorithms. Despite the rawness of the data (which was never collected for predictive purposes), the models have been successful, increasing mean time between faults (MTBF).
This is only the start of the process, though, as the resulting data has to be translated in order to make sense to industry experts and allow for integration into business systems to support accurate decision-making.
A Virtuous Circle
As SD-WAN vendors compete to create the best machine learning-powered networks, the scope for even more sophisticated machine learning applications (including those utilizing the IoT, virtualization, and other cutting-edge technologies) will only continue to grow. As these applications push the boundaries even further, the best SD-WAN networks will continue to evolve in kind to keep pace.
Here we see the iteration of a virtuous circle of symbiotic growth towards what has been termed the World of Intent, where we tell the technology what we want, and it just goes about delivering it in the most efficient way possible.
Conclusion
Any cutting-edge technology, whether it be distributed processing, the Internet of Things, or, for the purposes of this article, machine learning, is only as effective as the underlying network through which it draws its computing power. The development of these technologies will necessitate an equal improvement in SD-WANs or whatever other smart network technologies lie on the horizon. Both will continue to evolve symbiotically as global IT networks become ever more powerful, responsive, efficient and, dare we say it, intelligent.
Ben Ferguson is the Senior Network Architect and Vice President of Shamrock Consulting Group, the leader in technical procurement for telecommunications,AWS cost reduction, data communications,SD-WAN solutions provider, dark fiber procurement, and cloud services.
Since his departure from Biochemical research in 2004, he has built core competencies around enterprise-wide area network architecture, high-density data center deployments, public and private cloud deployments, and Voice over IP telephony.
Ben has designed hundreds of wide-area networks for some of the largest companies in the world. When he takes the occasional break from designing networks, he enjoys surfing, golf, working out, trying new restaurants and spending time with his wife Linsey and his dog, Hamilton.
Want to write an article for our blog?Read our requirements and guidelines to become a contributor.
