May 1, 2017
A recent article identified what Gartner considers to be the top 10 strategic technology trends for 2017, one of which is machine learning. Per that article, machine learning has reached a tipping point and will increasingly augment and extend virtually every technology-enabled service and application. While I agree with that assessment, I don’t think that machine learning is a well understood discipline. Thus, I will use this blog to discuss some of the basic concepts of machine learning and to describe how machine learning is likely to impact networking in the very near term.
An algorithm is a specified set of steps that is applied to a data set. Think of it as a recipe. In a traditional environment, an algorithm doesn’t change over time unless someone manually changes it. So much of our traditional approach to networking has been based on the use of an extensive set of algorithms such as Dijkstra’s algorithm. Dijkstra’s algorithm is used to find the shortest paths between nodes in a graph and it is the working principle behind link-state routing protocols such as OSPF and IS-IS.
Artificial intelligence (AI) is a branch of computer science that focuses on the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception and decision-making. Machine Learning is a subset of AI that focuses on the practice of using algorithms to parse data, learn from it, and then make either a determination or prediction about something. In contrast to a static algorithm such as Dijkstra’s, a critical aspect of machine learning is that the machine is “trained” using large amounts of data and algorithms that give the machine the ability to continually learn how to perform a given task.
The technical basis for AI and machine learning goes back over fifty years but until recently these disciplines didn’t garner a lot of attention outside of academe. It isn’t unusual for a concept to stay dormant for long periods of time until one or more components of the environment changes. In the case of AI and machine learning, some of the recent environmental changes that have enabled these disciplines to blossom include the growth in computer processing power along with the evolution of cloud computing and big data technologies.
AI and machine learning began to get popular attention in 1997 when an IBM system called Deep Blue won the world chess playing championship. This attention grew in 2011 when another IBM system called Watson won the TV show Jeopardy. More recently, a lot of popular attention has been paid to the idea of leveraging the use of AI and machine learning to create a self-driving car.
Facebook is a leader in the use of AI and machine learning. One example is that Facebook has developed a tool called DeepText which extracts meaning from the words users post by leveraging AI and machine learning to analyze the words contextually. At present the tool is used to direct people towards products they may want to purchase based on the conversations they are having. In addition, Facebook recently announced that it has deployed AI-driven technology to identify users at risk of self-harm and offer them support.
Amazon is another company that has invested heavily in AI and machine learning. In addition to using those disciplines internally, in late 2016 Amazon announced a set of services, called Amazon AI services. According to Amazon, “Based on the same proven, highly scalable products and services built by the thousands of deep learning and machine learning experts across Amazon, Amazon AI services provide high-quality, high-accuracy AI capabilities that are scalable and cost-effective.”
The over-arching impact that machine learning will have on networking is that it will represent a huge step forward for automation. One use case for machine learning is the automated detection and analysis of anomalous behavior. In the not very distant future, well-trained, machine learning based systems will be able to identify security risks and intrusions and will also be able to troubleshoot performance problems before they impact users. Another key use case that will likely be mainstream relatively soon focuses on the path selection functionality contained in SD-WAN solutions. One example of what you can expect to see soon is that a properly trained SD-WAN solution will be able to anticipate congestion on a given WAN link and automatically divert traffic to an alternative link.
AI and machine learning are no longer topics that are applicable just to academics. These disciplines are currently used in production for a broad range of use cases and that use will grow rapidly in the near term. As a minimum, network organizations need to work with their existing vendors to understand each vendor’s roadmap for introducing these disciplines into their products and solutions. As part of that activity, network organizations need to understand the use cases that their vendors are targeting and the timeline for when this new functionality will be in production.
In addition to querying their existing vendors about their use of AI and machine learning, network organizations that are in the process of implementing new functionality should use vendors’ plans for AI and machine learning as a decision criteria. For example, network organizations that are evaluating SD-WAN solutions should choose an SD-WAN solution in part based on the plans that the vendor has to add value by incorporating AI and machine learning into their solution.