In our previous blog article on “Developing a Successful End-to-End Complex Event Processing (CEP) Strategy,” we discussed how to use an Interconnection Oriented Architecture™ (IOA®) strategy to deploy CEP at the digital edge, allowing DevOps teams to fully integrate and realize the vertical and horizontal objectives of their business and IT organizations. Once in place, a CEP implementation allows you to leverage predictive algorithmic services based on integrated data flows within your digital platform to learn and understand emerging behavioral patterns, and recommend preemptive actions in response to potential threats or failures.
Predictive modeling relies on algorithms that have been trained to analyze historical data and apply common business rules to incoming data. By leveraging methods such as artificial intelligence, machine learning, and statistics, predictive algorithmic services can create new analytics models to forecast the probability of an outcome, allowing you to gain insights into your data, or eventually predict a shift in business activity.
The industry forces behind deploying predictive algorithmic services at the digital edge
Even with the best real-time business and IT process data and a deep understanding of how your CEP environment works (with controls in place), traditional operational responses are still largely reactive rather than proactive. In this dynamic environment, people in the response flow can add an hour. When everything is constantly changing, and you’re continuously pivoting, how can you accurately plan or forecast your business or anticipate an event-the holy grail of digital business?
The forces driving the deployment of predictive algorithmic services include:
Today’s businesses are rapidly changing. Processes are quickly shifting to digital. IT services are becoming virtualized and rule-based, as discussed in our blog on CEP, removing software development friction between business functions and IT. However, to build and apply operational intelligence to operating models (supported by complex event patterns), you need to introduce machine learning-based, predictive algorithmic services. For example, whereas CEP can tell you the water in the building is not working, predictive analytics can tell you the impact of that malfunction-by revealing how much water you will need for the super soaker fight schedule in the office at 4:00 PM.
One lesson distributed technology has taught us is technology breeds more technology. The challenge behind this is the types of information and volumes of data are changing and rapidly increasing. Predictive analytics helps us sort through increasing volumes of diverse data to mine information that is critical to the business.
When you put CEP together with predictive analytics, you can start modelling what you should do next with pre-determined actions and implications. Over time, experience builds into the models, which in turn improves the predictions. In short, it’s strategic planning at scale and speed.
The constraints behind deploying end-to-end predictive algorithmic services
Following are some of the challenges to deploying an integrated, end-to-end predictive algorithmic service at the digital edge of your business:
Once you have deployed CEP, you need predictive algorithmic services to provide the insights that have the greatest impact on your business. But you need an end-to-end process for integrating CEP and predictive analytics to realize that value. For example, CEP can correlate the indicators that may lead to CPU, memory, disk or network failures that can occur. However, only predictive analytics can help you calculate the meantime to failure (MTTF) at both the component and system level and send an alert when you’re approaching a MTTF incident in order to avoid a catastrophic outage.
In cybersecurity, there seems to be an infinite number of different unrelated and related events and user behaviors that you must consider to prevent an attack. CEP can help identify and gather those events, but predictive analytics will help you figure out which of these events matters so that you don’t go chasing every anomaly.
With the massive amounts of different date types flowing into businesses today, it’s almost impossible to mind the gems that provide the most business value. CEP can help you correlate the information to find “data diamonds,” but only predictive analytics can take diamonds in the rough and reveal the shining insights that your business needs to survive.
To properly address these constraints, your digital platform needs to apply both CEP and algorithmic services to start recommending-and then take preemptive actions to forecast outcomes and act upon them accurately with speed and scale.
The solution for deploying digital platform algorithmic services at the digital edge
The design pattern for deploying digital platform algorithmic services at the digital edge prescribes leveraging an IOA strategy, a framework for directly and securely interconnecting people, locations, clouds and data at the digital edge. Once you’ve implemented the digital platform analytic capabilities from the complex event processing design pattern, you can apply operational intelligence to your operating models (supported by complex event patterns) via machine learning.
Before looking deeply into the data science aspects, the solution here is to broadly apply inductive learning to the analytic models already on the platform. The outcomes will be a mix of predictions around data volumes and capacity with relationship discovery (upstream and downstream dependencies). Other predictions are related to business behaviors and the likelihood of outcomes defined today to improve targeting. The goal is to start modeling what you should do next with actions and implications already determined. Over time, experience is factored into the models, improving the predictions. The result is strategic planning at scale and speed.
Take the following steps in conjunction with your CEP implementation to deploy predictive analytics services within an IOA framework (see diagram below):
Implement machine learning services or connect selected SaaS provider(s)’s respective services (defined here as an “engine”) within a digital edge node, along with complex event processing capabilities.
Connect the engine to analytical models and gathered historical data.
Begin training the engine, modeling known good and bad conditions and improving accuracy by using known historical information.
Compare the engine’s model to the current model and dashboard and start testing predictions. If the predictions are wrong, there is more learning (or potentially other data) needed.
Move from binary decisions to detecting complex behaviors to regression predictions about optimal conditions and implications, as well as deterministic outcomes.
Start applying this learning (from the predictive algorithms) more broadly to meaningful situations and strategic planning, enabling you to learn and control digital business globally at scale.
Digital Platform Predicitve Analytics Design Pattern
The benefits of predictive algorithmic services
When integrated with the company’s CEP implementation, placing predictive algorithmic services at the digital edge provides the following benefits:
Models can be continually refreshed with updated dependencies, improving situational awareness and understanding.
Viewpoints on business activity, outages (e.g., failure predictions), utilization (e.g., waste), behaviors (e.g., fraud, etc.) and risk (e.g., operations heat map) can be built and continually refreshed.
A digital dashboard and various trends analysis reports are mostly automated and more accurate.
The platform is more results-oriented and will evolve and incorporate external business events and activity, with an eye toward digital business models and direction.
All aspects of the business have become digital, completing the digital platform.
Predictive algorithmic services use case
In the age of the digital economy, use cases for predictive analytics have become plentiful and are touching nearly all industries. And, there are literally hundreds of exciting applications – for example: assessing why patients are readmitted to the hospital or predicting the ideal spot for a retail store and customers’ buying potential. The analysis and the outcomes can help determine whether a patient needs mental health treatment in addition to treating a physical ailment, or they can assist in future product and service strategies. Today’s predictive algorithms can even emulate human skills such as sentiment analysis.
One of the most interesting examples of predictive algorithmic services is the field of preventive maintenance. In this application, usually large, high-value machinery assets are equipped with a series of IoT sensors that are transporting captured data across a network (often what’s referred to as the “Industrial Internet”) to an analytics engine.
For example, imagine an array of sensors fitted to a jet engine. These detectors can capture vast amounts of data related to engine revolutions, temperature, vibrations, fuel consumption, and number of full and reverse thrust instances. A common denominator for all predictive analytics is the ability to quickly access and distribute data, derived from various sources and often in the multiple terabytes, to the analytics algorithm to quickly produce the outcomes. So, it is crucial for a jet engine manufacturer to deploy multiple globally distributed digital edge nodes, each of which can receive and process the data from each region. This regional approach ensures optimal performance and rapid results.
These results help the jet engine manufacturer proactively schedule maintenance and dispatch parts and technicians ahead of the airliner’s next touchdown. The ability to offer this type of insight and proactive maintenance adds significant value to the manufacturer’s products and enables a new dimension of “service wrap.” And it’s a win for the airline, too, as the parts can be replaced and maintained before an actual fault occurs and causes costly delays and downtime or even worse, an accident.
From a business’s margin perspective, predicative analytics can also provide insights into more fuel-efficient flight paths, depending on wind speed, air traffic patterns and poor weather conditions. This enables an airline to react in real-time to changing conditions and deliver passengers to their destinations on time, an important competitive differentiation for companies in this extremely time and cost-sensitive transportation market.
In the next blog article, we’ll discuss placing predictive algorithmic services at the digital edge.
In the meantime, visit the IOA Knowledge Base for vendor-neutral blueprints that take you step-by-step through the right patterns for your architecture, or if you’re ready to begin architecting for the digital edge now, contact an Equinix Global Solutions Architect.
You also may be interested in reading other blogs in the IOA Application Blueprint Design Pattern series:
How to Localize Digital Services at the Edge for Greater Performance and QoS
Accelerating Digital Business by Deploying Application API Management at the Edge
How to Plumb your Messaging Infrastructure for Application Flows at the Edge
Developing a Successful End-to-End Complex Event Processing Strategy