It seems human beings have always felt the need to gain a more in-depth insight into their situations or questions about their future. We can get a glimpse into how strong that need is by looking at the number of different methods of divination (the attempt to gain insight using standardized, often occult or supernatural rituals). When looking into different types of divination, I expected some I was already familiar with from history lessons, like augury, numerology, I Ching, and maybe a few others. But what I did not expect was finding more than 300 different divination methods from different times in history and different cultures. Our need for more profound insight is indeed strong!
Today, in what I like to believe is the age governed by science rather than superstition, we often turn to data science to obtain insights and answers to questions about past, present, and even the future. Unlike methods of divination mentioned above, data science applies the scientific method when working with data to obtain knowledge by carrying out experiments or making an empirical observation.
The insights we desire in our networking domain have a wide range from why something happened. For example, when we are searching for a root cause of a problem that already occurred. Does any portion of our network need attention right now? Have we reached the limits of capacity? Do we need to plan a complete network redesign or just replace parts of our infrastructure to avoid future problems?
This can be something as simple as a database query or a simple search. For example: how many iPhones are currently connected to my network? Or we can compile simple statistics and perform statistical modeling. For example: what is the average link utilization at a particular site? What is the average link utilization in my network? Or we can use more advanced methods such as machine learning (ML) to make predictions. For example: What will the link utilization in my network be like next week?
When questions get more complex, we no longer just query the data, but use it to create a model that mimics reality in order get our answers. The more precise the answer needs to be, the more time we need to spend on improving those models and the common requirement: lots and lots of quality data. This means gathering enough data so we can capture required historical trends and granular data to capture all the nuances and of course, the correct data that contains data points that will help answer our question.
There is a saying attributed to a famous statistician George Fox George Fox: “All models are wrong, but some are useful”. This means that the models never really represent the real world fully but are rather a simplified representation and therein lies one of the reasons we decided to offer the choice of unlimited data to our ExtremeCloud IQ customers. For example, the anomaly detection models might need to be improved or rebuilt in the future because of new discoveries or better tools become available. This can be done quickly if the data is immediately available.
We can also always go back and explain the decision-making process of our ML/AI and how we built a specific model. Testing for repeatability of insights provided is essential for any science. We might also be interested in long term seasonality and trends. Or the question we are asking might be phrased differently (for example, what is the normal link utilization in my network? – becomes what is the normal link utilization at a particular site?). And finally, we might start asking questions we haven’t asked before. Unlimited data gives maximum flexibility to quickly and accurately find those answers.
Every enterprise network needs a copilot. ExtremeCloud™ IQ CoPilot provides explainable ML/AI builds trust by providing readable output of how insights were derived, enabling you to automate operations, enhance security and enrich user experiences with confidence. Please travel to this URL to learn more: https://extremeengldev.wpengine.com/uk/copilot/.
