Although artificial intelligence (AI) applications have an increasingly important role in automating routine IT operations, their use is becoming even more of a necessity for managing enterprise cloud domains.

Managing computer systems for any enterprise involves a wide variety of tasks, such as diagnosing problems, maintaining user accessibility, ensuring application performance, and detecting and neutralizing security threats, to name just a few of the most critical ones. AIOps, or artificial intelligence–driven operations apps, is a family of software that automates these and other activities.

The advent of cloud computing operations (CloudOps) has added new layers of focus in the operations sphere that in turn has created a requirement for a different branch of AIOps functionality that deals specifically with cloud activities. Some basic AIOps functions were overviewed in an earlier article, but AIOps for cloud functions are perhaps even more important for enterprises using distributed cloud environments.

Many IT departments already have individual tools that can perform one operations-related analysis task well (or several adequately), but these tools don't necessarily integrate with each other. Even collectively, an unmatched assortment of such tools usually provides a view of just a piece of overall cloud environments, which now extend into areas such as infrastructure-as-a-service and platform-as-a-service modes, in addition to the software-as-a-service realm that has been present for more than a decade.

AIOps Requirements for CloudOps

Although CloudOps functions can overlap with requirements for non-cloud operations, user expectations put a greater demand than before on user- and customer-service areas. These include such tasks as ensuring 24/7 operations, optimizing user access from any device type, streamlining communications traffic across networks, improving efficiency of troubleshooting task diagnostics, maintaining business continuity, enhancing user and customer experiences, meeting service-level agreement (SLA) and other contractual commitments, and lowering costs.

The convoluted nature of many multi-cloud infrastructures and the volume of transactions handled via the cloud provide such an overwhelming amount of data that with today's limited IT personnel budgets, complete analysis of one problem before the next five emergencies pop up is becoming nearly impossible. A 2020 survey of CIOs estimated that a typical mobile or web app may cross an average of 37 different components or technologies. Traditional IT tools that typically focus on a single aspect of application or system performance also can't keep up. In the face of these complexities, a turn to techniques of artificial intelligence (AI) and machine learning (ML) is rapidly becoming mandatory.

What's needed are a set of operations tools that can oversee any kind of cloud operations, monitor the vast amounts of CloudOps data coming in, summarize it into a UI for human engineers to view, and hopefully, use AI and ML techniques to learn to analyze it all to highlight problems and offer some solutions for heading off future problems.

Other useful attributes would be recognizing similarities between alerts from varying sources, identifying which alerts are the most significant from a large volume of them streaming concurrently, and correlating data to suggest root causes and problems and their solutions. In addition, AIOps tools should be capable of monitoring, if not running on, virtually any hardware or software platform.

I Spy with My Little Eye

Using most of these abilities together creates a state called observability. In CloudOps, the term implies not only the ability to view the status of any hardware and software components of a cloud system, or systems attached to it, but also the ability to correlate and contextualize many types of data to postulate root causes of problems and their possible solutions. By applying AI and ML algorithms to all incoming data, an AIOps app should be able to filter out the noise of minor and duplicate incident reports and alarms, analyze traces and log events, and identify deeper problems that may be generating alerts of different kinds in assorted contexts. Traditional tools are too limited to monitor such varied sources of information as, for example, cloud-resident data, system logs, and container contents. CloudOps personnel need a more comprehensive toolset.

AIOps apps are analysis systems that can take in the whole landscape and provide solution suggestions that go beyond mere error triage. These include application performance monitoring; data collection on the telemetry, user-action, and business-data fronts; analysis of problem root causes that takes into account multiple processes running in multiple locations; and tracing paths to find out exactly where an error originated, rather than simply pointing out the symptom the error provoked. AIOps apps strive to interpret meaning from all this data but without human help beyond initial setup and training of the AI.

AIOps solutions currently available can't do all this yet, partly because the field is new enough that there is no industry standard for what "any" AIOps app should be able to do. However, AIOps apps for CloudOps are reaching for the general goal of being not just an observability platform, but one that can use AI to help with analysis of complex systems.

Stepping Beyond User-Complaint Firefighting

Obtaining the best results from conversion to an AIOps system actually requires looking beyond the operational environment to make operations conform to the strategy of a particular enterprise, rather than just generically supporting IT functions well. This means incorporating into AIOps some kind of view of key performance indicators (KPIs) for the business itself.

For example, it's not enough to simply troubleshoot and prevent user problems. It's also necessary to provide insights into ways of improving as well as gauging application performance, and measuring network performance and preventing downtime incidents rather than simply reacting to them more efficiently. This process involves monitoring not only application usage but more traditionally nebulous metrics such as user satisfaction and employee productivity. Yet even that isn't fully sufficient.

By definition, KPIs also include such aspects as customer satisfaction, repeat business, lead generation, conversion rates, customer acquisition costs, and any number of enterprise-defined business goals. Automation and analysis can help corporate entities improve their bottom lines by making some non-operations business processes run more efficiently. How useful might it be to know some of this information as well as server uptime? What's more, cloud environments particularly need predictive maintenance augmentation to prevent access disruptions and outages.

This added dimension of functionality is why AIOps apps also help with the task of breaking down corporate silos to access a wider array of data to analyze. Enterprises can automate distributed application operations when AIOps and observability work together to provide the combined services of detection, analysis, and automation. The independent learning nature of AI also opens the door to IT operations that can manage, secure, heal, and optimize themselves once properly trained. The amount of data needed to provide sufficient insights to achieve this, while impossible for a person to fully grasp, isn't an insurmountable problem for an AI-based system that can generate greater understanding of an environment the more data it sees.

AIOps apps can act as a coordinator of all these aspects. They can forestall the problem of an enterprise's multi-cloud environment from becoming so large and complex that people can no longer efficiently control it. They also present the possibility of an IT operation that can cure its own disorders within an enterprise that is also gaining a better understanding of its customers and its business functions.

Cloud Security with AIOps

Data security is vital in enterprise environments, and AIOps can help with that as well.

Because AIOps apps' data-gathering includes information on the nature, severity, and impact of security incidents, they can also provide insights that will help IT develop better security intelligence and alert strategies. AI/ML methods can also assist in generating predictability models for types of security threats. Networks and endpoint functions can be better analyzed for signs of compromise. The ability of AI to conduct database searches, analyze social networks, mine text, and detect anomalies gives investigators new tools for detecting fraud.

In addition, AI apps trained to check external information sources about security threats, malware, and Internet protocol addresses known to be problematic can learn to detect malware and similar threats on their own. Trained AI apps can also monitor data stored in cloud environments and classify it as to pattern, content, and various types of metadata to detect anomalies and streamline data operations.

Preparing for AIOps Adoption

Because its impact is potentially so comprehensive, some groundwork must be in place to successfully adopt any AIOps app. It's not a simple accounting program, after all.

Arrangements—technical and political—need to be made to give the app access to whatever data is necessary in order for it to carry out all the functions enterprise interests may require. If it's not already in place, an "agile" IT infrastructure in which software rather than hardware controls and virtualization are at the center of cloud operations is recommended. These two aspects alone may require cross-department cooperation, weeks if not months of preparation, and a C-suite champion at the very least. The AI will have to undergo a period of training, during which it is fed operational data, analyzes it, and has the results checked by IT people. Expectations for results will have to take into account this delay.

Once trained, the app will need the ability to alter security policies, network configurations, user accesses, and other functions in order to automate self-repair capabilities. Those permissions will need to be approved and established. Ideally, some efforts will be made to encourage a corporate culture that is tolerant of information sharing between different entities within the organization, rather than maintaining a siloed culture of data feudalism. A culture that further stresses continuous improvement of data quality and quantity will be valuable. A re-evaluation of current data-collection means and practices is also likely to be important.

AIOps is still a developing area of expertise, and its market requirements are, at this point, as much recommendations as achievements. No available app today can do everything a mature AIOps package could possibly do, but the pieces are falling into place. What's important to remember is that adopting an app of this type is a matter of matching current capabilities to your enterprise's particular needs today, coupled with your own assessment of how capable the vendor you select will be in accepting input for new capabilities and acting on market standards that will continue to evolve. However, despite whatever shortcomings may exist in current versions of products available today, AIOps is a class of IT operations solutions that eventually will revolutionize how IT and cloud operations are done. There are definitely benefits to be had from today's versions of AIOps apps.

The following products and services may help you find a CloudOps solution that is superior to the methods your enterprise is currently using, or at least give you more information about specific AIOps capabilities available now.

Aruba AI

Cascadeo.io

Centerity Edge

CloudFabrix Robotic Automation Fabric

Dynatrace

IBM Cloud Pak for Watson AIOps

Instana Enterprise Observability

Lakeside Systrack

Larsen & Toubro Infotech Mosaic AIOps

LogicMonitor

ScienceLogic SL1

Splunk

Sumo Logic

Virtana Virtual Wisdom

vRealize Operations

Zenoss

John Ghrist

• Email
• Site

John Ghrist has been a journalist, programmer, and systems manager in the computer industry since 1982. He has covered the market for IBM i servers and their predecessor platforms for more than a quarter century and has attended more than 25 COMMON conferences. A former editor-in-chief with Defense Computing and a senior editor with SystemiNEWS, John has written and edited hundreds of articles and blogs for more than a dozen print and electronic publications. You can reach him at ghrist@comcast.net.