How In-memory Computing with HTAP Powers IoT

Posted by Lalit Ahuja

Widespread adoption of IoT has led to organizations pushing the boundaries of what’s possible using the technology. Monitoring of operational assets and IT/OT integration, while extremely valuable, has evolved into generating actionable insights and true autonomous machine-to-machine (m2m) interaction. From autonomous vehicles to predictive maintenance to remote patient monitoring to the metaverse, real-time is the key, requiring not only high-speed data aggregation, but also complex analytics on that same data as soon as the data is generated.

While advances in network technology have significantly reduced information flow latency, managing transactional data and analyzing it in real-time continues to be a major challenge. One strategy that is leading the charge to address this is Hybrid Transactional Analytical Processing (HTAP) powered by in-memory computing.

Two types of IoT processing – Analytical and Transactional

In the past, many IoT implementations and use cases were unidirectional. Streaming data coming in from all sorts of sensors and devices was pushed at high speeds to large data stores. This data was then analyzed using typical big data analytics technologies to draw relevant conclusions for appropriate human intervention.

For true m2m interaction, however, the communication must be bidirectional between sensors and devices, and the processing must be real-time to support decisions and subsequent actions – all in a matter of milliseconds.

Consider an example use case: managing an end-to-end protein manufacturing process at a biotech firm’s manufacturing plant using IoT. To enforce process quality control and avoid the possibility of losing an entire batch of protein, the system must continuously receive data from various sensors and controllers and analyze that data to make decisions and trigger control actions – all in real-time during the ongoing manufacturing process.

For the biotech firm, this bidirectional, real-time communication and action can save an entire batch worth millions of dollars. For other use cases, such as autonomous vehicles, the stakes are even higher. Lives are at stake, and a single millisecond delay could prove catastrophic.

HTAP: Simultaneous transaction processing and analytics

These types of “transactional” IoT use cases require not only writing or persisting data at high speeds, but also processing every single incoming piece of information, analyzing it contextually, making a decision and initiating an action, all in real-time.

This is where data stores with HTAP capabilities come into play. HTAP creates that centralized, highly scalable data storage and processing tier that can process transactions and also perform inline analytics on that rapidly changing transactional data. In the case of the protein manufacturing process, the data coming in from any sensor is contextualized with data coming from other sensors at that exact time, analyzed based on an intelligent anomaly detection model, for example, leading to an appropriate decision. This decision leads the IoT system to trigger an action that is then initiated on the appropriate part of the manufacturing process to bring it back within the overall process control limits.

This type of automated process control would not be possible without the aggregation and analysis of data, execution of some complex and intelligent anomaly detection model and then initiating an action based on that intelligence, all done in real-time within the transactional scope of that ongoing highly sensitive manufacturing process.

HTAP and in-memory computing

Currently, the most effective way to deploy HTAP is with in-memory computing. In-memory computing platforms are typically deployed on a cluster of commodity servers, either on-premises, in public or private clouds, or on hybrid architectures. By pooling the available memory and compute from across the cluster, the in-memory computing platform can store vast amounts of data in-memory and use massively parallel processing (MPP) to deliver up to 1,000x faster performance for applications that were built on disk-based databases.

Analyzing the data in the same in-memory computing cluster where it is being written eliminates the movement of data over the network between the traditional OLTP and OLAP systems. This is really the key to enabling real-time, “transalytic” processing and decision-making. Further, the computing cluster can scale horizontally to petabytes of in-memory data, and some in-memory computing platforms offer multi-tiered computing to allow seamless processing of data cached in memory or stored on disk.

As companies in a wide range of industries continue to explore the potential of IoT, the demand to implement real-time, bidirectional M2M use cases will soar. HTAP is a proven and cost-effective strategy for making the solutions to these use cases a reality.