For nearly a decade we have witnessed a rapid evolution of cloud infrastructures, applications, and operations. At first, the cloud was mainly about migrating legacy applications from on-premise infrastructures to cloud data centers. This was a key to taking advantage of the scalability, flexibility, and Quality of Service (QoS) of cloud infrastructures. In later stages, application developers had to revise the design and implementation of their applications to allow them to benefit from their direct deployment on the cloud. Over the years, application developers and cloud providers realized the benefits of unifying application development with cloud operations, which gave rise to the DevOps (Development and Operations) paradigm. The latter complements the agility of software development processes with flexible the (re)configuration of the cloud operations. In the scope of DevOps, application development and cloud operations are closely related and jointly optimized.
Nowadays, enterprises leverage large scale cloud computing infrastructures, which comprise elements from public cloud, private cloud, and on-premise data centre infrastructures. Furthermore, in modern deployments, the complexity of the underlying cloud infrastructures is completely virtualized thanks to container technologies and microservices. Specifically, most non-trivial cloud applications are comprised of a rich set of microservices, which can be deployed and scaled-up independently in different parts of the infrastructure. In this context, application developers and cloud infrastructure operators are provided with large volumes of data about the operation of their underlying cloud infrastructures and of their applications. In the years to come, the processing of these data will provide valuable insights regarding the operation of the cloud infrastructure. This will substantially increase the automation and intelligence of DevOps operations.
From DevOps to AIOps
The processing of large amounts of data from the cloud infrastructure enables enterprises to realize a shift from DevOps towards data-driven models for configuring and managing cloud infrastructures. One of the most prominent models for data-driven infrastructure management is AIOps, which is expected to help enterprises revolutionize their IT operations based on Artificial Intelligence (AI) technologies. AIOps is about using AI to process data from the infrastructure towards automating and optimizing cloud operations. As a prominent example, AIOps enables automated, AI-based detection of incidents, management of faults, and intelligent root cause analysis.
AIOps deployments assess, detect, analyze, and resolve incidents across mission-critical workloads over virtualized cloud infrastructures. The main components of an AIOps infrastructure include:
- Data collection modules: AIOps hinges on the processing of large amounts of data from the cloud infrastructures. To this end, AIOps platforms include a variety of data collection modules such as monitoring probes over logs and networked devices. Such probes ensure the availability of extensive, rich, and diverse data about the assets of the cloud infrastructure and the applications running over them.
- BigData modules: AIOps infrastructures include BigData modules that undertake the scalable persistence and management of the collected data. These modules provide the means for handling large amounts of data in diverse formats and with different rates of ingestion.
- Data Mining Modules: To extract insights about the context of cloud operations and to identify important patterns (e.g., faults and anomalies) AIOps platforms employ data mining techniques. Machine Learning (ML) models have a prominent position among these techniques. For example, deep learning techniques are used to identify or to predict abnormalities in the operation of the infrastructure. Likewise, ML-based recommender systems enable prescriptive analytics and provide actionable recommendations about how to configure the infrastructure for optimal QoS. The recommendations include for example suggestions about allocation of resources (e.g., memory) to specific application workloads.
- Actuation Modules: The main role of the actuation modules is to operationalize the recommendations on the cloud infrastructure. They are the actuators of the AIOps infrastructure, much in the same way the data collection modules are the sensors of the AIOps platforms.
AIOps Benefits
AIOps provide a host of benefits to cloud operators and application developers. Specifically:
- Timely Diagnosis of Issues: AIOps analyses data from all the different parts of the virtualized cloud infrastructure. This analysis facilitates the faster identification of problems and abnormalities, given that it obviates the need for human administrators to inspect logs. Furthermore, it enables the discovery of hidden patterns of failures i.e. patterns that can be hardly identified by humans.
- Accelerated Decision Making: In an AIOps environment, administrators are provided with actionable recommendations about infrastructure optimizations. These recommendations are provided in a faster and more automated manner.
- Separating Useful Information from the Noise: AIOps infrastructures provide powerful data filtering functions, which help to separate important information from the noise of the data collection process. This is a foundation for faster and improved business intelligence.
- Optimized Deployments: AIOps platforms operate over the full set of elements of a virtualized cloud infrastructure, including public clouds, private cloud, and on-premise data centers. This facilitates application developers to select the optimal deployment option based on a combination of performance, cost, automation, and security considerations.
- Trusted Analytics: In cases where AIOps leverage AI algorithms that operate as black boxes (e.g., deep learning analytics), it is possible to provide transparency regarding ML-driven decisions. For instance, emerging AIOps infrastructures provide the means for explaining AI models towards ensuring their transparency. This is key to increasing administrators’ and other human operators’ trust in the AIOPs analytics outcomes.
The AIOps Pipeline
AIOps functionalities are usually implemented as AI/ML pipelines over a vast amount of data collected from the virtualized cloud infrastructure. A typical AIOps pipelines consist of the following processing stages:
- Data Collection and Monitoring: In this step collected data are filtered to provide baseline monitoring functionalities. This helps administrators identify whether things are working well, or if some point requires attention.
- Analysis and Optimization: This applies more sophisticated analytics techniques (e.g., deep learning) to identify, predict, and anticipate interesting patterns of the infrastructure’s behavior. This includes for example the identification or prediction of a failure using predictive analysis. Following the identification of issues, the analysis focuses on producing actionable recommendations about how to best (re)configure the cloud infrastructure.
- Actionable Recommendations: This is the last step of a typical pipeline. It focuses on the operationalization of the recommendations through the actuators of the AIOps infrastructure.
In the next couple of years AIOps will be increasingly deployed to enhance the automation and intelligence of cloud operations. In the medium term, AIOps platforms will offer a rich set of monitoring and analytics functionalities, such as behavioral analysis, failure pattern matching, and predictive analytics. Enterprises must therefore consider the state of the art in AIOps platforms towards positioning themselves in the data driven infrastructure management landscape and planning their adoption steps accordingly.
