Automate Your Digital Infrastructure for Better Results – Interconnections – The Equinix Blog

• Automating physical fiber interconnection in data centers using robotics. This removes the potential for manual error and increases agility and flexibility by streamlining changes to customers’ interconnected ecosystems.
• Automating physical security and safety measures within data centers. Using AI for video feed analysis, data center operators can automatically detect and respond to anomalies like unauthorized penetration of the security perimeter or potential hazards like water or cardboard boxes on the data center floor.
• Intent-based networking, which represents the logical next step in the evolution of software-defined networking. Network administrators can define the desired outcome or business objective they hope to meet. The network then applies AI and machine learning capabilities to determine how best to achieve that intent, and adapts itself accordingly.
• Automating bare metal provisioning and management. One example of this is Tinkerbell, an open-source project built and maintained by the Equinix Metal® team. Tinkerbell uses microservices and an API-based approach to make it quicker and easier for businesses to stand up bare metal infrastructure when and where they need it.
• Automating compute container movement between data centers. Enterprises can automatically move their compute instances based on energy availability. This can help decrease their overall power costs, while also improving environmental sustainability by emphasizing renewable energy use.
• Automating cybersecurity threat detection. Human cybersecurity experts simply can’t scale their capabilities wide enough to perform manual threat detection. Automated systems are able to parse massive volumes of data to identify anomalies before they become an issue.

Understanding the basics of automation

To gain a better grasp of how automation can support digital infrastructure optimization, it may be helpful to take a step back and consider what automation is at its most basic level. In its seminal white paper on the topic, IBM defined the core components of automation using the MAPE loop: Monitor, Analyze, Plan and Execute.[1]

To truly be considered automated, a system must be able to perform each of these functions in sequence:

• Monitor: Collecting and aggregating data from a particular managed resource.
• Analyze: Using the data collected to model and predict potential future scenarios. Increasingly, enterprises are using machine learning and deep learning techniques in this phase to create models that can detect problems and also predict the future.
• Plan: Applying the insights gained in the Analyze stage to determine how to achieve specific objectives.
• Execute: Putting the plan into action.

The MAPE loop is a “loop” in the sense that it’s not a one-time process. Automated systems will consistently cycle through each of the four functions, allowing them to adjust as new data points enter the system. The objectives that enterprises hope to achieve through the MAPE loop will take the form of service-level objectives (SLOs) or service-level agreements (SLAs).

Furthermore, one usually creates a hierarchy of MAPE loops, where there are smaller MAPE loops corresponding to the smaller components of the system. To use a car analogy, there can be separate MAPE loops for the transmission system, air conditioning system, braking system, etc. These loops all interact with each other to satisfy SLOs corresponding to speed, temperature, safety, etc. In many deployments, it is prudent to initially have a human involved in the MAPE loop to certify that the machine-generated automation scripts make sense. After humans gain more trust in their models, they can then fully automate the MAPE loop.

In addition to describing the desired system objectives, SLAs also describe the penalty function if the SLOs are not satisfied. System architects will set different SLOs or SLAs for every aspect of their digital infrastructure, including compute, storage and networking. Certain objectives, such as performance and availability, will be common across all digital infrastructure. Others will be unique to one particular area, such as durability for storage and jitter for networking.

Why AI and black-box models represent the future of automation via digital twins

AI and machine learning sit at the heart of automation. Understanding the different machine learning models and how they’ve changed over time can be helpful in understanding why automation works the way it does.

Essentially, machine learning models are classified using two different criteria: how accurate they are and how easy they are for human observers to understand and interpret. Based on these criteria, we can separate models into white-box models and black-box models:

• White-box models rely on rules that are manually specified by experts. As such, it’s fairly simple to look at a white-box model and understand why it returns the predictive results that it does. However, white-box models tend to be less accurate, for the simple reason that experts aren’t always able to update rules quickly enough to keep up with the features being added in the system that is being modeled.
• Black-box models include more modern, sophisticated machine learning models, such as neural networks. They are called “black boxes” because they deduce rules or patterns without human intervention, which makes it more difficult for observers to understand how and why they make the predictions they do. Black-box models tend to be more accurate, because they are able to update themselves in real time to keep up with changes in the observed environment and the system being modeled via retraining the model.

One area where black-box models are helping drive better results in digital infrastructure is through the use of digital twins. Organizations are creating digital twins to represent cars, airplanes, factories, shopping stores, data centers, storage systems and more. Black-box models are ideal for digital twin initiatives because they remove the need for human intervention to support change management. This helps ensure the models are as accurate and resilient as possible, even as the circumstances surrounding the observed environment change over time.

By creating an exact digital representation of their infrastructure assets, enterprises can perform predictive analytics to accurately forecast how their infrastructure would perform under specific scenarios. These analytics insights can be applied to optimize for a number of different objectives, including performance, resilience and energy-efficiency. Based on the predictive insights, automation models can intervene and make corrections anytime a system is in danger of not meeting its SLAs. Enterprises deploy digital twins close to where the data is generated (e.g., on factory floors and in shopping stores) and they pose queries on these models using real-time data.

Thanks to the growing availability of black-box machine learning models, we now find ourselves in an exciting new era of automation. Increasingly, users are able to specify their objectives, and then let the system figure out how best to meet those objectives, with no manual intervention required. This “intent-based” approach provides almost limitless potential to redefine what enterprises can accomplish with their digital infrastructure, and Equinix is proud to help our customers make the most of what automation has to offer.