Artificial intelligence (AI) and machine learning are two of the latest networking buzzwords being thrown around the industry. The problem is many enterprise network managers remain confused about the real value of these vastly useful technologies.
Emerging network analytics services, powered by AI and machine learning promise to transform traditional infrastructure management models by simplifying operations, lowering costs, and giving unprecedented insights into the user experience – improving the productivity of both IT professionals and their users.
For network staff, the concept and value of these technologies is extremely powerful if applied to the right problems.
Good problems to have for AI
One big problem is today's operational challenge in dealing with the mass of user, device, application and network service data traversing the enterprise access infrastructure. Machine learning, if applied properly, is an ideal solution for making sense of all this data to figure out how all the different parts of the network are behaving with each other.
A second big problem is the need to automate the network within a grand closed loop. The use AI and all this "big data" is key to making this happen. But first, the industry must get the ‘making sense of the data’ part right among many other things.
Today, network managers must wade through volumes of data from Wi-Fi controllers, server logs, wired packet data and application transactions, analyzing and correlating all this data to determine the health of network as well as trends and patterns of network behavior across the stack that impact user performance. Then, they manually apply changes to the network with no real way to definitively determine whether those changes worked or not.
Conventional network management and monitoring tools, never designed or developed to deal with these 21st century realities, are ill-equipped to automate this process.
Enter AI and machine learning.
First things first
Simply put, artificial intelligence is the development of computer systems able to perform tasks that normally require (super) human intelligence.
Rather than forcing people to perform increasingly complex calculations from a variety of data sources, work in AI has concentrated on mimicking human decision-making processes and carrying out tasks in ever more human ways to enable more predictive problem solving.
Related to this, machine learning is an application of AI. It is a toolkit of algorithms that provide systems the ability to automatically learn and improve from experience without being explicitly programmed to do so.
The process of learning begins with observations of data, and looking for trends, patterns and anomalies within the data to make increasingly better correlations, inferences and predictions. Machine learning software “learns” by discovering the processes that generate the observed outcomes of particular inputs. Finally, machine learning provides a framework to make predictions and recommendations as to what will improve the overall system.
Theory is great. What now?
So how can all this magic be usefully applied, in a practical way, to help IT and network staff drive down costs, drive up productivity and deliver better user experience on the network?
Machine learning is the ideal tool to automate many of the traditional infrastructure management processes that are performed manually. Specifically, in the context of enterprise access networks, it:
- Eliminates costly and cumbersome manual analysis and correlation of myriad network data sources by network staff,
- Identifies specific and systemic user network performance problems across the entire IP stack and makes recommendations and predictions on fixing them,
- Delivers a single source of network truth that can be used by different factions within the network team, each responsible for their own services,
- Minimizes the finger-pointing among IT staff when issues arise, and
- Predicts potential network problems and capacity requirements before they happen.
Because machines, not people, are staring at every client network transaction 24 hours a day, network managers care able to determine who, what, when, where and why network problems are occurring – and what do about them – even if they don't know where to look or what questions to ask.
Cisco, HPE, Mist and Nyansa, the top talkers in the market giving lots of lip service to the use of AI and machine learning. Nyansa, the only of the four with a pure-play commercial offering of machine learning for access networks. Its Voyance network analytics platform provides a good glimpse into what can be practically achieved through the technology's application.
Putting ML to (net) work
Machine learning is useful but only when fed tons of relevant data. On the Enterprise access network, that includes live packets off the wired network, wireless metrics from WLAN controllers, SYSLOG data from different network servers, ad other network data Sources. Machine learning is used to quickly analyze all this different data, correlating it across different network layers. This is something that's not practically possible with people trying to manually correlate it.
The beauty of these machine learning solutions is that they can be used without server agents, client software or intrusive architectural changes – using the data already running over the network.
Central to machine learning is the use of massively-scalable cloud computing resources, sophisticated big data repositories and analytics algorithms that turn everything into meaningful and understandable actions that network managers can take.
Once analyzed, this data is distilled to surface trends and patterns impacting the performance of every device on the network. The resulting insights, not clearly visibility or easily achieved by network managers, tell IT staff exactly where, when and why user connectivity falters.
Because every client network transaction is analyzed by machines, pinpointing precisely where the network is struggling and quickly be determined.
Are issues occurring on a specific VLAN? In a specific location? Is the problem with a certain Wi-Fi access point or group of APs? A certain type of device? Is it an application problem? DNS or DHCP issue? For a given group users? What are some concrete actions I can take to improve DNS experience in my network? Without machine learning, getting answers to these questions can take days or even weeks.
The right time at the right place
Given the invasion of new data now hitting enterprise access networks, machine learning and AI couldn't be more welcome technologies for taking the pressure of network managers to do more with less.
While AI is simply a general term describing automating manual or complex tasks, machine learning is a toolkit of algorithms that enable automatic learning from ‘big data’ already running over today's networks.
Armed with these technologies, network managers can now better understand where they have issues with user experience, get recommendations of actions to take and, ultimately, automate the configuration and operation of the infrastructure. This is network nirvana by almost any definition.