Machine Learning (ML) is one of the most researched topics in computer science. It’s been around for decades. However, many people consider it just another buzzword, or even worse, confuse it with Artificial Intelligence (AI). However, the two are not the same.
Machine Learning is the science of a machine learning and improving without being specifically programed to do so. Artificial Intelligence is the base technology that makes this possible. Think of ML as a subset of AI. It’s important to keep in mind that all Machine Learning is Artificial Intelligence but not all Artificial Intelligence is Machine Learning.
More than clarifying the difference between the two, it’s important to understand why Machine Learning has gained so much attention over the past few years.
Several factors have contributed to the growth of this discipline:
- ML makes digital feel more human. For example, Twitter uses a complex algorithm to create your timeline. This means that no two users will have the same experience — even if they follow the same people or have the same followers.
- ML keeps getting better. When’s the last time you used Alexa, Siri, or Cortana? If it’s less than a few months ago, try again. You’ll be pleasantly surprised.
- ML is available to more developers. All cloud providers have offerings in the field, including Google, AWS, and Azure. Check them out!
Movere is helping customers maximize their investment in digital technology by helping to wrangle the massive amounts of data generated every day. Whether it’s understanding what’s running in an existing data center, or how your on-premises licensing translates to a cloud environment, we have found that change is the only constant in IT. When you think about how you measure change, and plan for it in a consistent way, data is the best place to start. Nobody can argue with good data as a predictor of what’s to come. However, we produce too much data and can’t consume it at the same pace; definitely not in its raw form. So how do enterprises leverage data when we all seem to be drowning in it?
This is where ML comes to the rescue
In the past few years, both the computing power required to analyze vast amounts of data, and the storage needed to capture it, has gotten significantly less expensive. This means that the power of ML can be made available to more users, at a much lower cost, reducing the barrier to entry for everyone. What is missing is a model that can be developed, tried, and tested before one starts to benefit.
First Step: De-noise the data
Let’s use a practical example of what noise means. Take the CPU utilization of a server. This is a classic metric that DevOps use all the time.
Here is the same server, looking just at one day. We focus on a Tuesday since it seems that was the busiest day.
Tuesday could have been the busiest day because:
a) Monday was a public holiday
b) Everyone’s most productive day is Tuesday
c) Random software running in the background
d) All of the above
The answer is D, all of the above. The point is humans infer assumptions, machines don’t. And when you apply ML to the task of understanding what is happening, it looks more like this:
This particular server is actually doing very little work most of the time. The exceptions are Tuesday, Wednesday, and Thursday (real user traffic > 1 percent). How did we figure that out? By simply de-noising the data. Once you clear out the noise (hint: we used ML) and look at actual process level data (think Chrome.exe running at 20 percent vs. your whole server running at 20 percent), along with network traffic and other variables that isolate real activity from system activity, you can make the logical assertion that this server is actually only used three out of seven days.
Next step: Only pay for what you use!
When planning a cloud migration, it is crucial to know that you are sizing optimally before you move (for more on this, see Optimization before Migration), but also after you’ve made the jump.
After looking at billions of data points from devices that are running all sorts of apps and workloads, you start seeing patterns that emerge which can be used to predict when a server should be on at full capacity, on at reduced capacity, or completely off. Long gone are the days when only Dev and Test servers should be powered off after 5 p.m. Most, if not all servers can be powered off during certain time windows; the question is when? It’s not possible to make those predictions when looking at metrics from the macro level through a single pane of glass. However, with the help of machine learning you can. And with the money you can save from shaving hours and days off your bill, the investment you make in machine learning will pay off quickly.
Leveraging machine learning is the single most efficient way to turn all your collected data into actionable insights that lead to effective business decisions.