You’ve heard it over and over: data is (digital) gold in today’s world. The more data you have, the more opportunity it brings to anyone or anything that knows how to properly mine and use it. It’s being consumed and translated by the millions every single second – learning your shopping and browsing habits to suggest similar items from the latest seasonal trends, recommending television shows or movies based on what you stream, to knowing the type of food you eat so restaurants can appeal to your changing tastes. Understanding what a person or thing is searching for and then presenting a recommendation is pretty straightforward. But in reality, it’s extremely difficult to extract and analyze meaning from large volumes of big data. Translating it to make an intelligent decision within a reasonable timeframe? Let’s bump that complexity up a few notches.
Data analytics and AI / ML
There are different categories of data analytics and machine learning (ML) methodologies that help businesses and organizations harness this power to gain a competitive edge. Let’s start with the three main categories for data analytics:
- Descriptive analytics – What has happened?
- This method uses data mining and data aggregation to analyze or monitor clusters of data to summarize what has happened.
- Predictive analytics – What will happen?
- This method examines historical data and events to forecast or predict what may happen in the future.
- Prescriptive analytics – What will I do?
- This method optimizes data and runs through hundreds or thousands of algorithms to find the best possible outcome.
Let’s take it a step further and explore ML, which is a subset of artificial intelligence (AI). To quickly differentiate, AI is the action of making a better decision or solving a problem. ML is the process of using different algorithms and neural networks to support AI in making that decision or solving that problem.
Like data analytics, we can also break ML down into different categories that help AI improve a situation or problem:
- Supervised learning
- This is broken down into two subfields: classification and regression. Classification identifies A or B, yes or no – the output is categorical. Is the email genuine or spam; is that credit card transaction legitimate or fraudulent? And regression uses X and Y axes to determine a relationship between a feature, number or variable with known data. Based on certain zip codes, school rankings and square footage, for example, a regression model can predict how much houses will sell for with precision.
- Unsupervised learning
- Unlike supervised learning, where you have known sets of data, this model takes unstructured and unlabeled data to group or correlate clusters of similar data together. When you send in a saliva sample to do DNA sequencing, there is no data on your background, where you came from or your family’s health history. By clustering similar DNA genomes from thousands of other samples, this method can accurately predict your origins, the health issues linked to those origins and your life expectancy.
- Reinforcement learning
- This method starts with data collection, or observing. It then progresses to (continuous) learning to make an action that returns the largest reward. An example is GPS navigation systems. Not the basic one that comes with your car where it maps Point A to Point B, but the modern mobile GPS apps that may give you a different route to work every day of the week based on traffic conditions that are changing by the minute. This model continuously checks drive times of other users on the same commute, looks for traffic accidents and road closures, factors in when school zone traffic is in effect, assesses weather conditions and so on – and uses this data to find the largest reward factor, that is, the fastest route to your destination. And it continuously checks the route throughout your drive to check if any unforeseen circumstances pop up that warrant a re-route.
Self-tuning with vRealize AI
The modern datacenter and its application workload performance demands is just like driving on a Monday morning in San Francisco – extremely hard to predict. You have noisy virtual machines throughout the day that drive at different speeds, VDI virtual machines that require extra read throughput on the morning commute and then the SQL database virtual machines that need extra write throughput in the evening commute. And just as you’d use a GPS app to guide you through traffic, we are introducing VMware vRealize AI beta that continuously learns and optimizes vSAN tunables so your business applications are running at peak performance.
As you might already be familiar with vRealize Operations, one of the key tenants of the self-driving data center is ‘Continuous Performance Optimization’ and this is where you’ll navigate to enable the vRealize AI service for vSAN. VMware vRealize AI learns about your operating environment and adapts to changing dynamics, ensuring optimization for the desired KPI – selecting all or specific vSAN clusters that you need to apply the self-tuning and automatic optimization to.
Using reinforcement learning, vRealize AI collects data and metrics from vCenter and vSAN, stores the target data in the SaaS data lake, monitors the data sampling of vSAN I/O and resource utilization, analyzes performance to determine what actions to make based on continuous learning results, and then dynamically self-tunes vSAN and vSphere parameters to give you the best performance results. In a never-ending continuous closed-loop, vRealize Operations and vRealize AI provides visualization of your latest performance index readings, searchability for a specific time frame to compare your current KPIs with and without vRealize AI enabled.
If you are running VMware vSAN and/or vRealize Operations, then I invite you to learn more about vRealize AI. And if you have dynamic workloads running on your vSAN clusters that need performance improvement, please sign up for the beta program.
For any other questions or more information, please don’t hesitate to email us at: email@example.com