IT operations teams must keep up with the speed of DevOps and continuous delivery while aligning IT capabilities with the business. Analytics can play a significant role helping you meet that goal.
Before you can improve monitoring and resolve the issues that IT monitoring uncovers, however, you’ll need a foundation in AI operations analytics (AIOps). The goal of analytics in IT operations is to help you avoid problems in the first place by analyzing the enormous volume of data generated by your core IT systems, networks, security mechanisms, and distributed devices.
Analytics takes on massive quantities of data filtered through various combinations of preconfigured algorithms, machine learning, and trained analysts, and continually tunes all the tools that offer AIOps capabilities. When well-tuned tools spot anomalies in log data, analyst teams ideally can get to the heart of the problem quickly.
These four modes of analytics that will help your teams understand what happened, why it happened, whether it might happen again—and what you can do about it.
The what and the why
These two modes of analytics go hand-in-hand. They’re as basic as cause and effect, but in reverse order. You usually know the effect first (log files show that a server is down) and the cause second (the power was interrupted). So the “what” and the “why” can be considered the fundamental building blocks of your team’s analytics capability.
What happened: Descriptive analytics
The simplest form of descriptive analytics is a report, which can come in the form of a text message, printout, physical gauge, or automated dashboard (a virtual gauge) registering some level of operational capacity or system health.
If these mechanisms indicate that all is working within normal parameters, great. But when something goes out of spec, knowing that as soon as possible is obviously the key to fast correction.
Why it happened: Diagnostic analytics
When you describe your sore throat symptoms to a doctor, you’re simply saying what you know is wrong. When the doctor swabs your throat, then says you have a streptococcus infection, she is performing a diagnosis—i.e., explaining why you feel bad. (Sorry if all this sounds overly simplistic. It’s just that many of us confuse the “what” and the “why” all the time.)
Diagnostics is the critical step in getting to a solution to the problem. In this case, the diagnosis leads the doctor to put you on a course of antibiotics.
Will it happen again? And can you prevent it?
When you move from “what and why” into this next pair of analytic modes, you move from past tense to future tense. That is, descriptions and diagnostics are about events that have occurred in the past, which usually require some remedy.
On the other hand, predictive and prescriptive analytics provide insights to help you prevent those events from being repeated in the future.
Will it happen again? Predictive analytics
Knowing what is normal is the only way to know when something is not. But getting to a sense of normal takes time. The goal of machine learning is to compress that time, so predictions based on analyzed data can lead to faster decisions.
And that means lots of data. The more information you have about what causes problems, the better equipped you are to prevent them. Once you know the indicators that spell your organization’s unique trigger points, you can make rational decisions as those conditions occur again. Tools can alert you to the conditions; what you do about the “prediction” is up to you.
Let’s keep it from happening: Prescriptive analytics
Knowing how to prevent health problems is a long-held goal in medicine. The same idea applies to IT Ops. Finely tuned algorithms that “learn” over time can suggest, for example, the optimal configuration of virtual machines based on factors such as workload, performance, location, and power consumption.
The idea is to learn everything about what has gone wrong before, in order to suggest remedies that prevent future problems.
IT is a complex beast
Torrey Jones, a consultant at Greenlight Group and analytics practitioner, offered a historical perspective on the rise of analytics.
The past few years have brought several technologies to the mainstream that have rendered previously machine-unreadable datasets now readable, he said. This opens new ways to handle data and for how it gets interpreted, aggregated, correlated, and consumed by machines. All of which allow new ways to make this data meaningful for human consumption.
That explains the growth of analytics as an area of interest for many IT disciplines, including IT operations.
Head off your next headache
One you understand these four modes of AIOps you’ll be ready to put them into practice, leveraging machine learning, root-cause analysis, and behavioral analysis in IT operations analytics tools to head problems off at the pass.