Chapter 8: Architecting AIOps

In this chapter, we will learn how artificial intelligence for IT operations (AIOps) platforms are designed and what sets them apart from any other monitoring tool. You will get a better understanding of why these platforms will become a necessity in the future of IT. The chapter starts by explaining the logical architecture model and then drills down to the key components in AIOps using data lakes and analysis through machine learning. We will define the reference architecture for an AIOps platform and learn how this will drive operations as we integrate the technical services architecture in AIOps. The chapter will provide insights into algorithms, anomaly detection, and auto-remediation.

After completing this chapter, you will be able to recognize the various key components of AIOps and define the reference architecture. You will learn some important lessons in implementing AIOps in enterprises, avoiding some major pitfalls.

In this chapter...

Understanding the logical architecture

Before we dig into the architecture of AIOps, we need to understand the structure of logical architecture. In the previous chapter (Chapter 7, Understanding the Impact of AI to DevOps), we learned that one of the first steps to get started with the implementation of AIOPs is to get full visibility of all our IT assets and processes. This is a requirement to feed the AIOps model with the basic information of how the IT environment looks. Most AIOps tooling will scan the environment using agents, but that is not sufficient. It also needs to understand the relationship between assets and what process flows are implemented. That is covered in the logical architecture.

The logical architecture is not about technology. It doesn't care what type of machines are used or what software. Logical architecture describes systems without the definition of the underlying technology. It describes the relationship between logical components in a system...

Defining the key components of AIOps architecture

In this section, we will discuss the key components of AIOps. Then, we will look at the operating services that provide input for AIOps, and finally, in the last section, we will draft the reference architecture.

First, let's recap why we should have AIOps in IT and specifically DevOps. IT has become more complicated. Systems are hosted on various platforms, connecting to other systems, using a lot of different data sources, with an equal amount of variety in data formats. We have come to the point where it has almost become impossible for a human to maintain an overview of these complex landscapes. Yet, the market is demanding new features at a rapidly increasing pace. Developers are under a lot of pressure to deliver code that fulfills new functionality, while operators need to make sure that systems are running stably. AIOps can help you do the following:

• Process and evaluate data: AIOps can rapidly, in almost real...

Integrating AIOps with service architecture

So far, we have looked at the logical and different components of AIOps. One of the main reasons to bring AI into operations is to relieve operators of a lot of manual tasks by proactively monitoring systems and even mitigating possible risks before they actually materialize. In other words, AIOps is invented to reduce what Site Reliability Engineering (SRE) calls toil. That's the whole point of implementing DevOps and AIOps: to reduce toil and create time to continuously design and build better systems.

But it's not only about the systems themselves. Enterprises also have processes to deliver services. That's the domain of the service architecture. Or, better and more precise: the technical service architecture. That includes the processes that are discussed in the following subsections.

Monitoring

This is not something that simply comes out of the box. On the contrary, architects need to define a monitoring architecture...

Defining the reference architecture for AIOps

In the previous sections, we studied the logical architecture of systems, the components of AIOps, and the technical service architecture. All these building blocks are used to define the architecture for AIOps. In this section, we will look at the reference architecture for AIOps.

First, let's recap the goal of AIOps. In Chapter 7, Understanding the Impact of AI to DevOps, we discussed the Key Performance Indicators (KPIs) for AIOps:

• Mean Time to Detect (MTTD)
• Mean Time to Acknowledge (MTTA)
• Mean Time to Resolve (MTTR)

AIOps adds artificial intelligence to IT operations, using big data analytics and machine learning (ML). The AIOps system collects and aggregates data from various systems and tools, in order to detect issues and anomalies fast, comparing real-time data with historical data that reflect the original desired state of systems. Through ML it learns how to mitigate issues by automated actions...

Summary

This chapter was a deep dive into AIOps. This is a rather new domain, but very promising. We've learned how AIOps platforms are built and learn as they are implemented in enterprises. It's important to understand that you need a logical architecture to have a complete overview of how systems fulfill functionality and how they are related to other systems, without already knowing the full technical details of these systems.

Next, we defined the key components of AIOps, being big data and machine or deep learning. AI only performs if it has access to enough relevant data on which it can execute analytic models. These models will teach the platform how to detect issues, anomalies, and other events, predict the impact on the IT landscape, find root causes faster, and eventually trigger actions. These actions can be automated. AIOps platforms will avoid a lot of tedious, repetitive work for operators, something that is called toil in SRE.

We've learned what...

Questions

1. What does the presentation layer in logical architecture do?
2. AIOps platforms are able to detect deviations from expected system behavior through algorithms. What do we call this process of detecting deviations, a key feature of AIOps?
3. AIOps works with operational system data and data coming from events such as incidents and problems. What is this latter type of data referred to by Gartner?
4. False or true: data cleansing is essential in AIOps.

Further reading

Pragmatic Enterprise Architecture, by James V. Luisi, 2014