Create the Next Generation of Streaming Analytics

A Breakthrough for Introspection of Streaming data

Traditional stream-processing and complex event-processing systems focus on extracting patterns from incoming telemetry, but they can’t track dynamic information about individual data sources. This makes it much more difficult to fully analyze what incoming telemetry is saying. For example, an IoT predictive analytics application attempting to avoid an impending failure in a population of medical freezers must look at more than just trends in temperature readings. It needs to evaluate these readings in the context of each freezer’s operational history, recent maintenance, and current state to get a complete picture of the freezer’s actual condition.

That’s where the power of real-time digital twins comes in. While digital twin models have been used for several years in product life cycle management, their application to stateful stream-processing has only now been made possible by advances in scalable, in-memory computing. Unlike traditional streaming pipelines, like Apache Storm and Flink, real-time digital twins offer a simple, intuitive technique for organizing important, dynamically evolving, state information about each individual data source and using that information to enhance the real-time analysis of incoming telemetry. This enables deeper introspection than previously possible and leads to significantly more effective feedback — all within milliseconds.

Real-time digital twins also provide a powerful means for deploying machine learning (ML) capabilities that track incoming telemetry and look for anomalies that require alerting. By running within digital twins, ML algorithms can be tailored for each type of device and its parameters, and they can run independently and simultaneously for thousands of data sources.

Equally important, the state-tracking provided by real-time digital twins allows immediate, aggregate analytics to be performed every few seconds. Instead of deferring aggregate analytics to batch processing on Spark, real-time digital twins enable important patterns and trends to be quickly spotted, analyzed, and handled. This dramatically improves situational awareness. For example, if a regional power outage takes out a group of medical freezers, precise information about the scope of the outage can be immediately surfaced and the appropriate response implemented.

Wide Range of Applications

Real-time digital twins can enhance the ability of any stream-processing application to analyze the dynamic behavior of its data sources and respond fast. Here are just a few examples:

• Intelligent, real-time monitoring: fleet tracking, security monitoring, disaster recovery
• Financial services: portfolio tracking, wire fraud detection, stock back-testing
• Internet of Things (IoT): device tracking for manufacturing, vehicles, fixed and mobile devices
• Healthcare: real-time patient monitoring, medical device tracking and alerting
• Logistics: real-time inventory reconciliation, manufacturing flow optimization

Real-time digital twins enable real-time streaming analytics that previously could only be performed in offline, batch processing. Here are a few examples:

• They help IoT applications do a better job of predictive analytics when processing event messages by tracking the parameters of each device, when maintenance was last performed, known anomalies, and much more.
• They assist healthcare applications in interpreting real-time telemetry, such as blood-pressure and heart-rate readings, in the context of each patient’s medical history, medications, and recent incidents, so that more effective alerts can be generated when care is needed.
• They enable e-commerce applications to interpret website click-streams with the knowledge of each shopper’s demographics, brand preferences, and recent purchases to make more targeted product recommendations.

An Example in Fleet Tracking

Consider the use of real-time digital twins to track the movement of vehicles in a nationwide car or truck fleet. Each twin can track a specific vehicle using specific contextual information, such as the intended route, the driver’s profile, and the vehicle’s maintenance history; ML algorithms can continuously examine engine and cargo telemetry with predictive analytics. These twins can then alert dispatchers or drivers when problems are detected, such as a lost or erratic driver or impending maintenance issue with a vehicle. In additional, real-time aggregate analysis can detect regional issues affecting several vehicles, such as weather delays and closed highways. By boosting situational awareness, real-time digital twins enable dispatchers to quickly hone in on problems and react within seconds.

Deeper Introspection for Better Responses

Traditional CEP and stream processing pipelines, such as Apache Storm and Flink, are “stateless,” lacking knowledge about the dynamic state of each data source to help interpret incoming telemetry. Real-time digital twins overcome this limitation by tracking state information for each data source, opening the door to much deeper introspection and more effective responses in real time. These twins can incorporate algorithmic code, rules engines, or even machine learning to help perform their analysis of incoming events.

Create Simulations that Scale

Enable Predictions and Improve Decision Making

Digital twins have historically been employed as a tool for simulating the detailed behavior of a complex physical entity, like a jet engine. The ScaleOut Digital Twin Streaming Service takes digital twins in a new direction: simulation of large systems. In addition to using real-time digital twins for streaming analytics, its highly scalable, in-memory computing architecture enables it to use digital twins to simulate many thousands of entities and their interactions. This provides a powerful new tool for extracting insights about complex systems that today’s managers must operate at peak efficiency. Its analytics and predictive capabilities enable better predictions and improved decision making,

Developers can use simulations in two primary use cases:

• to implement a workload generator for validating real-time analytics code in simulation prior to deployment in a live system, and
• to build a time-driven simulation model of a large system with many interacting entities for use as a tool for analysis and prediction.

Both use cases require a simulation platform that can host a large population of digital twins that represent simulated devices or other data sources and deliver high throughput. With its scalable, in-memory computing platform running on a cluster of servers, the ScaleOut Digital Twin Streaming Service provides just that.

Build a Workload Generator for Real-Time Analytics

Digital twins can implement a workload generator that generates telemetry used in validating streaming analytics code. Each digital twin models the behavior of a physical data source, such as a vehicle in fleet, and the messages it sends and receives. When running in simulation, thousands of digital twins can then generate a telemetry workload that feeds streaming analytics, such as a telematics application, designed to track and analyze its behavior. In fact, the streaming service enables digital twins to implement both the workload generator and the streaming analytics.

Here’s an example of using digital twins to generate a scalable workload for real-time analytics in a telematics application.

The diagram above shows a live system in which real-time digital twins track a fleet of trucks; each twin analyzes telemetry from a specific truck. It also shows a simulation in which digital twins simulate the behavior of trucks and generate telemetry messages. They feed them to their corresponding real-time digital twins to test and evaluate real-time analytics. The simulated trucks can be parameterized to model various behaviors, such as a lost truck, mechanical issue, or an erratic driver. Once the real-time analytics has been thoroughly evaluated in simulation, it can then be deployed in a live system.

Using digital twins to build a workload generator enables investigation of a wider range of scenarios than are likely to be encountered in typical, real-world use. Their parameterizable, stateful models of physical data sources help evaluate the ability of streaming analytics to analyze and respond in many possible situations.

Implement a Large Simulation Model

Digital twins also can implement a time-driven simulations model containing large groups of interacting physical entities. For example, an airline simulation might need to model passengers, aircraft, airport gates, and air traffic sectors to assess the impact of weather delays and other outages (such as ground stops). Digital twins maintain state information about the physical entities they represent, and they can run code at each time step in the simulation model’s execution to update their state.  They also can exchange messages to model interactions.

Here is a depiction of an airline tracking simulation with digital twins that implement aircraft airports, and passengers:

Simulations like this help systems managers plan logistics, evaluate decisions, and identify problems. They also can model live scenarios and help make predictions about the effects of decisions, such as airline schedule changes.

Easily Build Applications

The Next Generation in Streaming Analytics

Digital twins used for streaming analytics, called “real-time digital twins,” both simplify the design of stream-processing applications and improve the quality of analysis. The traditional approach relies on partitioning application code into multiple pipeline steps and using ad hoc techniques to access caches or databases.  This adds complexity and puts the burden on the developer to ensure fast performance.

Real-time digital twins sidestep this complexity by offering a simple, straightforward model for processing incoming telemetry based on tracking each data source’s dynamic state. This avoids the need to build streaming pipelines, and the execution platform automatically ensures high throughput and fast response times. The use of well understood, object-oriented development techniques further simplify the design process.

A Breakthrough for Simulation

Digital twins simplify the design of simulations, especially for systems with many interacting entities. Because a digital twin embodies both code and state information using an object-oriented software architecture, application developers can easily separate the design of simulation components from the overall orchestration of the simulation itself. This lets developers focus on application-specific issues while the ScaleOut Digital Twin Streaming Service manages concerns such as event processing, message delivery, load-balancing, scaling, and high availability.

Besides just keeping it simple for application developers, digital twins also enable ScaleOut’s in-memory computing platform to transparently scale so that it can handle very large numbers of simulated entities. For example, an airline simulation can model thousands of aircraft and hundreds of thousands of passengers:

What is a Digital Twin?

Digital twin models can take many forms. To simplify application design and enable transparent scaling, ScaleOut’s digital twin model uses a simple, object-oriented software design that supports both real-time analytics and simulation. Each digital twin comprises a state object holding dynamic state information and an application-defined, message-processing method and/or machine learning algorithm that analyzes incoming events and generates outgoing messages and alerts, as depicted in the following diagram:

When used to implement streaming analytics, real-time digital twins focus on analyzing incoming event messages to provide immediate feedback to their data sources (e.g., devices) within a live system. As event messages flow into the ScaleOut Digital Twin Streaming Service, a digital twin is created for each unique data source to process incoming messages from that data source. The message-processing method uses information in the state object to help analyze each event message, decide what action to take, and update its state. It can send a message back to the data source and/or send an alert if further action is required. (Some incoming messages may take the form of commands, which can be forwarded to the data source.)

When used to implement simulation entities, digital twins also contain a second method that processes events from the execution platform representing increments of simulation time. In this way, they can evolve simulation state at each time step and trigger the exchange of messages to model interactions.

The ScaleOut Digital Twin Streaming Service can simultaneously process incoming messages from many thousands (or even millions) of unique data sources in a live application, and it forwards each message to its corresponding real-time digital twin. In simulation, it can host many thousands (or even millions) of digital twins that represent simulation entities.

In addition, the streaming service can perform aggregate analytics and continuous queries across all digital twins by extracting information from the state objects, combining this information, and presenting the results in various types of charts and graphs.

Building Applications with Digital Twin Models

The ScaleOut Digital Twin Builder™ software toolkit enables developers to define object-oriented state information and analytics code for tracking telemetry from each type of data source (for example, a wind turbine or a fire alarm) or for tracking simulation state. This toolkit provides APIs in Java, and C# for constructing digital twin “models,” which are then deployed to the ScaleOut Digital Twin Streaming Service with just a few clicks in its web-based UI. The ScaleOut Model Development Toolkit™ gives analysts the ability to develop twins with easy-to-use business rules instead of code, and it allows machine learning algorithms to be deployed with no code required.

Each model defines the properties to be stored in the state objects and the user-defined analytics code and/or machine learning algorithms needed to process incoming telemetry or simulation time events. Once deployed, the ScaleOut Digital Twin Streaming Service uses these models to create unique “instances” of digital twins for all data sources or simulation entities as it processes incoming event messages.

Familiar, object-oriented class definitions in C# and Java simplify the development of advanced analysis algorithms and leverage everything developers already know about object-oriented programming. Equally important, they ensure a clean separation between application-specific code and the platform’s orchestration of event processing. The net result is that applications are straightforward to write and run without the need for specialized knowledge of complex APIs or platform semantics.

The following diagram depicts a streaming analytics application in which real-time digital twin instances analyze incoming telemetry from cars in a large rental car fleet. Each instance could hold detailed knowledge about a car’s rental contract, the driver’s demographics and driving record, and maintenance issues. With this information, the application’s message-processing method could, for example, alert managers when a driver repeatedly exceeds the speed limit according to criteria specific to the driver’s age and driving history or violates other terms of the rental contract, thus providing new insights on telemetry received from vehicles that otherwise would not be available in real time.

An application can define multiple real-time digital twin models to process telemetry from different types of devices. For example, an application which is analyzing telemetry from the components of a wind turbine might define three real-time digital twins corresponding to different components of the wind turbine, such as blades, generator, and control panel. Each component could send telemetry to three different digital twin instances, one of each type, as illustrated below:

Fast Deployment to the Cloud or On-Premises

The ScaleOut Digital Twin Builder™ software toolkit simplifies the development of Java, C#, and rules-based digital twin models by providing object-oriented classes that serve as a basis for defining these models. The next step is to deploy the models to ScaleOut’s cloud service or on-premises using a web-based UI. Once deployed, these models process incoming event messages, as illustrated below:

When used for streaming analytics, the ScaleOut Digital Twin Streaming Service’s UI lets the user easily connect the streaming service to numerous popular messaging hubs, including Microsoft Azure IoT Hub, Amazon AWS IoT Core, Kafka, and a REST web service, with more connectors to be released soon. Live data sources send event messages to a connected hub, and these messages are forwarded to ScaleOut’s streaming service. Once authenticated, the service receives incoming event messages and delivers them to their corresponding real-time digital twins. It also sends outgoing messages from twins back to their corresponding data sources using the connected hub. Connections to messaging hubs employ transparent scalability to maximize stream-processing throughput.

Maximize Situational Awareness

Aggregate Analytics for Situational Awareness

Real-time applications and simulations that track thousands of data sources, such as vehicles in a fleet or sensors on a security perimeter, need to be able to immediately identify hot spots and determine whether a pattern exists. This enables fast, strategic responses to emerging threats and optimal use of scarce resources.

Consider a cyber-security alerting network for a regional power grid. This network may comprise many thousands of software agents distributed across servers and controllers in the network, each assessing logins and commands to determine whether a potential attack is occurring. To respond effectively, it’s critical to immediately determine how widespread an attack is and where resources should be applied to contain it.

Digital twins provide a powerful software architecture for meeting this challenge. The software agents can send telemetry to digital twins which track the reporting status of all servers and controllers, including a current assessment of the likelihood of an attack. Real-time aggregate analytics can be performed across all the state objects for all twins to determine the overall  status of the network and, in the case of an attack, which locations have the highest threat assessment. This enables managers to see the complete situation and formulate a fast, strategic response.

No Programming Required

The ScaleOut Digital Twin Streaming Service’s UI enables fast, easy creation of real-time, aggregate analytics that combine the state of all real-time digital twins of a given type and provide immediate, graphical feedback. Each analytics “widget” displays as a bar, pie, or line chart and updates every few seconds with continuous, real-time results for both streaming analytics and simulations.

Each aggregate analytics operation is easily specified through the UI by selecting the parameters for a continuous “MapReduce” calculation (similar to an Excel pivot table), including:

• the real-time digital twin model
• the property within the state objects to be aggregated
• the aggregation operator (average, min, max, count)
• an optional property used to group the results
• the chart type (bar, column, pie, line)

The streaming service runs this operation every five seconds to update the contents of the chart with the latest values from its digital twins. This ensures that users always have the latest information on aggregate state of their digital twins.

In addition, the streaming service’s UI offers powerful query capabilities on the dynamic state of its digital twins, whether running in a live application or in simulation. Queries run in parallel across all servers to instantly identify which data sources have properties of interest. For example, in the above cyber-security application for a power grid, a query could be used to determine which specific nodes in the network have the highest current risk of attack based on recent network behavior. Queries can be configured to run continuously and plot their results on a geospatial map that maximizes situational awareness.

Because users can aggregate and query state information curated by digital twins instead of just viewing raw telemetry, they gain much deeper insights on the state of the system they are monitoring. Digital twins ingest and analyze telemetry to dynamically create state information that both filters out transient behaviors and surfaces important patterns and trends that need attention. This ability of digital twins to perform real-time analytics within milliseconds gives users a significantly more accurate picture of the situation and enables them to react more quickly and effectively.

Powerful Execution Platform

The ScaleOut Digital Twin Streaming Service deploys an application’s digital twins on a highly scalable, in-memory computing platform in the cloud and runs both stream-processing and aggregate analytics using an integrated compute engine. This delivers fast response times for devices (typically 1-3 milliseconds), and refreshes global statistics every 5 seconds.

ScaleOut’s in-memory computing platform is based on ScaleOut StreamServer and represents more than a decade of development and refinement. Its industry-leading software architecture offers several key benefits for hosting real-time digital twins:

• automatic correlation of incoming event messages by data source for delivery to their corresponding digital twin instances
• fast event delivery and response using an internal protocol based on the Reactive Extensions APIs and designed to minimize data motion
• fast access to state objects using scalable, highly available, memory based storage
• highly available event processing in conjunction with the connected message hub
• integrated MapReduce execution designed to minimize data motion
• scalable event-processing throughput by transparently adding cloud-based servers to keep event processing and aggregate analytics fast as the workload grows