With an increasing number of connected, distributed devices, there has been a gradual shift in how data is processed and analyzed. The trend is also based on the growth of emerging technologies, such as the Internet of Things (IoT), microservices, and event-driven applications, which influence the development of real-time analytics.   

This Refcard dives into how event stream processing represents this evolution by allowing continuous data analysis in the modern technology landscape. 

Event stream processing (ESP) encompasses a set of tools and practices used to analyze and make decisions based on continuous data streams. Unlike traditional data analytics, ESP is modeled on an event-driven architecture that continuously gathers and processes data points (events), making it a highly efficient, scalable framework to process big data in real-time.   

Similar to a microservice architecture, event stream processing utilizes services that share a common forum to perform tasks and interact with minimal shared knowledge of host services. Event stream processing implements this tell, don’t ask approach using a sink-and-source topology model. This can be explained further as:  

• A Source is an agent that generates and publishes an event.   
• A Source Connector acts as the link between the source and the ESP processing application.  
• A Sink acts as the storage platform to retain and exchange processed data.  
• A Sink Connector is used by the Sink to exchange processed data with other sinks or processors.  
• Sources and sinks are not directly linked; instead, the ESP processing application runs on a Stream Processing Engine (SPE).   
• Sources’ data are ingested and published into the storage layer without knowledge of the intended sinks address, and vice versa, which enables decoupling in data-driven applications.  

Event stream processing using a sink-and-source topology 

Evolution and Industry Adoption 

Due to increased demand for IoT, smart devices, and big data processing, cloud-based applications and emerging technologies are factors that have fueled the growth of event stream processing. As of 2018, event stream processing had a market size of $690 million, which is expected to grow at an annual rate of 21.6% to reach $1,838 million by 2023. Out of all the industries, though the trends have moderately fluctuated over the last several years, the largest adopters to date continue to be the banking, finance, and insurance sectors that leverage ESP to influence growth through real-time data analysis.   

Event Stream vs. Batch Processing 

Traditionally, applications employed batch processing to process data volumes in groups within a specified time interval. Such a model relied on a framework where data had to be retrieved, stored, and then processed/analyzed for future actions. Apart from this, because data was retrieved and stored in batches before analysis, batch processing systems did not allow efficient debugging of application errors. As a result, such systems were considered inefficient for the amount of time required to process data, as well as their inability to derive analytics in real-time for business functions where time is a major factor.   

On the contrary, event stream processing ingests, processes, and analyzes data streams in real-time. To do so, the ESP processor uses in-memory processing, pre-stored queries, and efficient analysis algorithms that allow data to be processed immediately the moment it is received.   

Apart from the basic differences above, there are other fundamental differences between how event streams and traditional batch processing operate. These include:  

• A batch processing system does not immediately push incoming data into the processor, thereby making it incapable of real-time processing. Event stream processing, on the other hand, relies on passive queries to instantly analyze incoming data to enable continuous, real-time data analytics.   
• At its elemental level, batch processing systems operate on disk storage data such as a filesystem or relational database. When processing data, I/O functions from disk storage prove to be an inefficient, time-constraining model, particularly known for the significant time lag. On the contrary, ESP leverages efficient in-memory processing by utilizing system RAM, flash memory, or in-memory databases that allow high IOPS for faster data processing. 
  

When to Use

Batch processing involves the collection of data sets over time, which are then pushed to the analytics system in batches. This makes it particularly useful for large volumes of static data that are generated and collected over a longer period of time. Batch processing is considered best where the results of processing analytics are not needed in real-time, and the requirement is to use a simple, efficient system that automates analytics to improve data quality.   

Some popular use cases of batch processing include:  

• Billing information generators  
• Payroll processing systems  
• Data processing systems that frequently work offline  

Event stream processing analyzes streaming data instantaneously, making it useful for high-velocity applications that generate events frequently or a business model that relies on real-time data analysis.  This makes ESP perfect for agile, time-critical applications such as:  

• Stock market operations  
• Air traffic control  
• Automated teller machines  
• RFID-based identification 
• Vehicle sensor systems  

Benefits of Event Stream Processing

ESP solutions act as critical enablers for organizations that rely on large amounts of continuous data to derive business decisions. As technology continues to evolve, applications are designed to churn out a wide variety of data per second, including server logs, application clickstreams, real-time user behavior, and social media feeds. While processing distinct data efficiently is one of the key features of ESP, the following are some of the benefits organizations achieve while embracing an ESP framework:  

Analysis and Processing of Big Data

ESP enables data to be captured and fed into analytics for instant processing and results. This is achieved by utilizing in-memory processing and continuous computation to process and analyze data constantly flowing through the system without limiting output time. Additionally, by leveraging in-memory processing, organizations can process enormous amounts of data instantaneously without the need to provision large data storage systems.   

Real-Time Decision-Making 

Events denote data points, while a stream denotes the continuous delivery of those events. By creating attributable data points, ESP makes it possible to visualize and present the incoming streams of data in real-time. This data can then be displayed on interactive dashboards or passed down to other processors for real-time decision-making. As a result, ESP applications are considered suitable for traffic monitoring, fraud detection, incident reports, and other functionalities that rely on instantaneous decision-making.  

Continuous Event Monitoring 

The ESP framework is built to analyze large volumes of data flowing through multiple data sources continuously. Every time a system’s state changes, the ESP records an event. The stream processing engine applies algorithms to aggregate event-based data, revealing key trends so that administrators can detect patterns and identify errors.   

Ultra-Low Latency   

One of the primary goals of stream processing is to allow real-time data analytics and processing for highly distributed, high-velocity applications. Besides this, as the architecture is built to adjust to changes in data patterns without affecting latency, ESP is considered ideal for applications requiring sub-second latency for workstreams, such as surveillance, robotics, and vehicle automation.   

Streaming SQL 

By using Streaming SQL, an ESP architecture enables a continuous query model for real-time analytics on a continuous stream of events. This allows organizations to utilize the declarative nature of SQL to filter, transform, aggregate, and enrich streams of rapidly changing data.   

Complements Edge Computing

ESP emphasizes real-time data collection through an adapter at the source processor. With the tell, don’t ask approach, ESP eliminates the need for a central, shared database. This reduces a round-trip to a server/database/cloud, enabling faster and advanced analytics for edge devices.  

To enable real-time data intelligence through event stream processing, organizations must enable a framework that allows the following capabilities:  

• High-speed data enrichment  

ESP emphasizes the need to ingest and process data continuously, thereby requiring high-speed filtering, sorting, and processing. To design an efficient ESP framework, organizations should take advantage of in-memory processing and distributed grid architectures that ensure low latency IOPS and quick enrichment of data.  

• Achieving sub-second latency   

When large, continuous data streams are involved, even the smallest delays impact critical time-based business decisions. It is required that the organization’s systems ingest, process, and respond to data with minimal latency to improve visibility and decision-making. 

• Ingesting multiple data types  

To derive analytical insights from a range of applications and devices, an ESP framework must be able to extract data from various data sources and then apply unified algorithms to present them in a uniform format. Additionally, the ingestion layer should support data transported over different protocols such as Kafka, MQTT, HTTPS, and AMQP.  

• Data scaling with microservices  

Modern devices and networks generate gigabytes of event data every second, requiring a hyper scalable infrastructure to enable processing. This can be achieved by leveraging event-driven microservices that allow organizations to keep up with rapid growth and data changes without having to reconfigure infrastructure since they enable highly scalable, multi-platform deployments.   

Elements of an Event Stream

Modern, agile applications are designed to consistently generate data for analytical insights. For applications that are built around event-driven architecture, data is streamed as a collection of data points, commonly referred to as events. These events typically denote a change in system states, such as user actions, device sensor outputs, or application statuses. The continuous generation of data points by application users results in a string of events known as an event stream. An event processing network is a collection of agents that transform, enrich, and validate events from producers then direct them to consumers.   

Event Streaming Architecture Components 

An end-to-end ESP stack consists of an assembly line of various platforms designed for specific functions such as storage, analytics, and processing. The stack offers developers a blueprint to design the overall workflow of configured components in the event-streaming data path. There are four main components that shape an event streaming architecture:  

Stream Processor

The stream processor, commonly referred to as the stream processing engine, is the agent that extracts data from the source. Once data is captured, it is translated to a standard message format, which it streams continuously. These processors rely on a hyper-performant messaging platform to enable massive streaming capacity and data persistence. Though such brokers mostly focus on streaming, these also apply queries to events from message queues and generate results. Some popular stream processors in use today include the Hazelcast Platform, Apache Flink, Apache Spark Streaming, and Confluent ksqlDB.

ETL  

ETL (Extract-Transform-Load) is the series of functions that prepare data for specific downstream actions by integrating data through three stages: extraction, transformation, and loading. ETL is the capability in an ESP system used to process large amounts of raw data from multiple sources into an enriched, consolidated view. This enables easier analysis and reporting, provisioning of historical context and business performance, and the maintenance of an accurate audit trail for data analytics and reporting. While ESP focuses on real-time data analytics, the ETL stages put data into context, which gives developers and administrators an all-around understanding of business trends and their underlying raw data.  

Query Engine 

Once the data goes through the ETL stages and is ready for consumption, the query engine processes and analyzes this data further to provide business insights. Event stream processing utilizes a list of passive queries to analyze and sequence data for storage or use by other processors. Some popular query engines that enable real-time data analytics for event streams include:  

• Elasticsearch  
• Amazon Athena  
• Amazon Redshift  
• Cassandra  

Data Storage 

Certain use cases require organizations to store streaming event data. While many ESP applications only work within a local control loop, some applications also require events to be recorded by default. To do so, organizations can extend traditional IT storage devices and databases to enable the post-processing of events. Some data storage options for event streams include:  

1. Data lakes — A data lake option is affordable and flexible since there’s no need to tabulate data, making it agile and highly efficient for enormous volumes of data. Data lakes, however, typically experience high latency, which presents a challenge for real-time analytics.  

2. Databases or data warehouses — Primarily used to store data for analysis that most commonly support SQL queries.  

3. Message brokers — Requires no additional components as it uses the inherent query engine for storage and is, therefore, the easiest to set up and scale.   

Though there are varied use cases of event stream processing, applications based on a streaming data architecture typically fall into the following two categories:  

Streaming Analytics 

This involves the processing of fast-moving, incoming data streams by performing simple, real-time calculations. By triggering continuous queries, streaming analytics enable the ability to constantly calculate, monitor, and manage live data streams. As compared to traditional analytics, streaming analytics-based apps process enormous amounts of data per second, and then seamlessly integrate the processed data into an application workflow or an external database. As these applications function on a framework that processes data before getting stored in databases, streaming analytics offer accelerated decision-making based on deeper insights of data visualization.   

Streaming analytics-based applications are typically used in:  

• Healthcare for the real-time monitoring of patient state  
• Financial markets for consistent market watch and transaction processing  
• In-home security for smart protection systems  
• Supply chain and logistics  
• Fleet operations   
• Identifying attack vectors to mitigate security incidents  

Modern applications generate consistent data streams that offer valuable business insights and historical trends. As a result, use cases of event stream processing continue to find favorable adoption in industries that churn out massive amounts of data. Some popular use cases for ESP include:

Internet of Things (IoT) Analytics 

As an increasing number of devices are connected to the Internet, there has been a consistent rise in data volumes that are collected, processed, and analyzed to unlock deeper insights. To help with this, by allowing the real-time processing of data generated by these devices’ sensors, ESP offers a continuous predictive analysis model that analyzes in-motion data to accurately predict future events.   

In contrast to traditional analytics, event stream processing enables real-time identification of system faults and predictive asset maintenance based on performance data. In such cases, each IoT endpoint acts as a data source that consistently generates event streams, which is then ingested and processed through the ESP framework to generate business analytics. ESP is also considered as the preferred solution for network optimization and maintenance of connected devices in power grids, traffic systems, smart cars, banking systems, and similar real-time IoT processing applications.   

Machine Learning 

Machine learning automates the building of data analysis models, and continuously updates these models to accurately predict future trends. With the increasing complexity of web applications and browsers, user actions generate events at high frequency, making most platforms data-centric. To help with this, ESP can be used to ingest, process, and extract immediate actions from the algorithms used in implementing machine learning frameworks, offering an improved user experience. ESP also helps build systems that can ingest new data streams in real-time and instantaneously update the analytical models for machine learning. Some popular frameworks for event processing in machine learning include The Hazelcast Platform (ML inference), Apache Spark, Hadoop, Apache Storm, and IBM InfoSphere Streams.  

Threat Modeling

By offering a lightweight, rapid approach for runtime verification of susceptible open points, stream processing enables the timely identification of application and infrastructure vulnerabilities. By deriving and following a robust threat model, ESP helps migrate security in distributed ecosystems by continuously monitoring threat actors in real-time. To do so, monitoring and processing a continuous stream of system events are cross-referenced with historical and predicted user behavior.  

Apart from the above, stream processing is widely popular in a range of use cases including payment processing, fraud detection, personalized user recommendations, and change data capture.   

Event stream processing allows modern data pipelines to be built around business value, ensuring competence in analytics and data processing. While doing so, ESP solutions abstract the complexity of traditional data structures into a single, simple platform that transforms a sequence of events into a stream of data. By adopting ESP, organizations can decouple the data pipeline from the compute elements, creating data pools that are easier to manage. As a result, organizations can focus on activities that add business values, while the ESP solution automatically manages the data plumbing.  

While there is a range of benefits of an efficiently designed ESP framework, organizations must also be wary of the challenges of implementing it, particularly with respect to storage and memory. As a closing thought, it is equally important that organizations emphasize the need to diligently design the right architecture, utilize the right tools, and follow best practices to enable an efficient, high-volume, low latency stream processing model for real-time analytics and business decisions.  

BMC Blogs. 2020. What is Stream Processing? Event Stream Processing Explained. [online] Available  at: <https://www.bmc.com/blogs/event-stream-processing/> [Accessed 17 June 2021].  

Confluent.io. 2020. Event-Driven Microservices Architecture. [online] Available at: [  <https://www.confluent.io/resources/event-driven-microservices> [Accessed 17 June 2021].  

Hazelcast. 2021. What is Event Stream Processing? How & When to Use It | Hazelcast. [online]  Available at: <https://hazelcast.com/glossary/event-stream-processing/> [Accessed 17 June  2021].  

Market, E., 2019. Event Stream Processing Market by Solutions, Services, and Application - 2023 |  MarketsandMarkets™. [online] Marketsandmarkets.com. Available at:   

<https://www.marketsandmarkets.com/Market-Reports/event-stream-processing-market 157457365.html> [Accessed 17 June 2021].  

Oppong, T., 2021. What Does Latency Mean for Stream Processing?. [online] AllTopStartups.  Available at: <https://alltopstartups.com/2017/03/15/what-does-latency-mean-for-stream processing/> [Accessed 17 June 2021].  

Rivery. 2021. Batch vs. Stream Processing: Pros and Cons | Rivery. [online] Available at:  <https://rivery.io/blog/batch-vs-stream-processing-pros-and-cons-2/> [Accessed 17 June  2021].  

Scalyr. 2021. Event Stream Processing: An Introduction and Helpful Guide | Scalyr. [online]  Available at: <https://www.scalyr.com/blog/event-stream-processing-guide/> [Accessed 17  June 2021].  

Snowplow. 2019. Market Guide for Event Stream Processing | Snowplow. [online] Available at:  <https://snowplowanalytics.com/resources/market-guide-event-stream-processing> [Accessed 17 June 2021].  

Upsolver. 2019. 4 Key Components of a Streaming Data Architecture (with Examples) | Upsolver.  [online] Available at: <https://www.upsolver.com/blog/streaming-data-architecture-key components> [Accessed 17 June 2021].