MQTT, once the cornerstone of IoT communication, is starting to show its age.

The Internet of Things (IoT) has reshaped entire industries, from smart homes to industrial automation, revolutionizing how we connect with the world. At the heart of this transformation has been MQTT, the trusted backbone of IoT communication for over a decade. But as AI pushes intelligence to the edge and devices grow smarter, MQTT is starting to buckle under the weight of these new demands.

This blog highlights conversations with dozens of IoT companies that have faced MQTT challenges, or embarked on a new mission, to revolutionize AI capabilities at the edge. We’ll conclude on how InfinyOn’s data streaming platform is an answer to these challenges.

Challenges with MQTT for IoT

For the past few years, we have interviewed dozens of IoT companies and have seen the following challenges surface over and over again:

• Resolve Intermitent Connectivity Issues
• Deploy AI Features at the Edge
• Transition to Real-Time Data Streaming
• Address Scallability and Operational Challenges
• Ensure Platform Flexibility

Let’s dive into each of these challenges.

Resolve Intermitent Connectivity Issues

IoT systems used by cars, scooters, vessels, medical devices, drones, robots, and other devices connected via cellular networks often encounter intermittent connectivity, resulting in a loss of connection. During these disruptions, critical data is at risk of being lost, mainly because MQTT lacks native caching capabilities. When a TCP/IP session is broken, messages sent with the Fire and Forget quality of service (QoS 0) are irretrievably lost. While this is acceptable in some industries, it can be a significant issue in others.

One approach to mitigate these issues is to deploy lightweight databases like InfluxDB or TimescaleDB on edge devices. These databases can store data temporarily until the connection is restored. However, it’s important to note that this approach requires additional development effort and introduces new challenges. Another is implementing custom compression and caching mechanisms to store data until connectivity is restored temporarily.

While these approaches are effective, they present significant challenges when deployed across tens of thousands of devices. Scaling these solutions can be a complex and daunting task, with challenges ranging from managing a multi-product stack to dealing with limited resources on edge devices.

• This stack uses multi-purpose databases that are not optimized for the edge, resulting in high latency and CPU utilization.
• The Ops team must manage a multi-product stack, which makes updating and monitoring complex.
• Edge devices have limited resources, which requires continuous resource distribution between services all these services.

Sometimes, companies outsource their edge networking to third-party providers that connect 4G and 5G networks to TCP/IP. We were surprised to find that even providers specializing in MQTT recommend customers transition to Webhooks for guaranteed message delivery.

Deploy AI Features at the Edge

Innovation in artificial intelligence has rapidly accelerated, with AI models increasing in capability by a factor of 1,000 every three years. This surge in demand for AI is also transforming the Internet of Things (IoT) landscape. A new trend is emerging: distributed AI, which involves training at the core and performing inference at the edge.

This transition is prompting organizations to reevaluate their IoT infrastructure. Many companies adopt a dual-stack approach, where MQTT is used for message passing, while an in-house solution or a vendor product manages the deployment and management of AI models. Others are looking for off-the-shelf solutions that integrate both message passing and AI model distribution within a single platform.

Furthermore, distributed AI extends beyond IoT, as organizations enhance data operations by moving AI inference closer to data sources to save money and improve performance.

These evolving requirements are set to transform IoT infrastructure well beyond MQTT’s capabilities.

Transition to Real-Time Data Streaming

The demand for real-time streaming in IoT applications spans across various industries where immediate insights and rapid decision-making are crucial. In smart cities, traffic management systems, public safety networks, and energy grids rely on continuous data to optimize operations and respond to emergencies.

• Autonomous vehicles and fleet management use real-time data for split-second decisions and route optimization, while healthcare devices stream patient vitals to ensure timely interventions.

• In industrial automation, factories monitor equipment and adjust workflows dynamically, and supply chains track shipments and inventory in real time to prevent delays and stockouts.

• Retailers leverage smart shelves and sensors to personalize promotions and manage inventory efficiently, while smart home devices adjust settings automatically based on user behavior or environmental changes.

• Agriculture relies on streaming data to optimize irrigation and crop health, and environmental sensors monitor air and water quality to respond swiftly to potential hazards.

• Telecom operators use real-time streaming to monitor performance, allocate resources, and mitigate security threats as they arise.

MQTT was designed as a lightweight communication layer, making it ideal for low-bandwidth environments that require reliable message delivery. However, as the volume of data generated by IoT devices continues to grow, MQTT struggles to handle large data streams efficiently. MQTT’s push-based model creates ongoing challenges in managing and analyzing data reliably. In contrast, streaming technologies, where clients control message delivery, provide a solid foundation for complex event processing and analytics.

Address Scalability and Operational Challenges

As described in the previous section, organizations must deploy AI inference on edge devices to reduce costs and enhance performance.

However, transitioning AI to the edge triggers a chain of new operational challenges. AI inference models are updated more frequently than device firmware, requiring a new deployment infrastructure layer. The results from AI inference feed into data models to compute aggregates that must adhere to specific schema objects. When schema objects are updated, they must remain compatible with the cloud services expected to ingest these aggregates. Therefore, an additional orchestration layer is necessary to ensure consistent management of these components.

MQTT lacks built-in capabilities for orchestration beyond simple message-passing rules. Some enterprises are developing this new orchestration layer in-house, while others are working with third-party companies such as InfinyOn to cover these changes.

Ensure Platform Flexibility

As edge computing and AI continue to transform the IoT landscape, platforms must support legacy systems while seamlessly integrating emerging use cases without requiring significant changes. They should effectively manage small-scale deployments and scale up efficiently without major overhauls. In addition to data ingestion, companies are seeking advanced pre-processing capabilities to handle large volumes of data and generate aggregates for integration with databases, data warehouses, cloud services, AI/ML services, and analytics tools.

MQTT lacks the flexibility required to address the evolving needs of IoT. It does not provide native support for advanced edge computing, real-time data streaming, or built-in AI capabilities. Furthermore, it does not include support for analytics or end-to-end orchestration with cloud services.

Conclusion

As AI advances in the IoT sector, embracing a modern platform that inherently addresses these challenges is essential, as well as minimizing costs, time delays, and the extra overhead of managing and integrating various technologies.

InfinyOn Solution

InfinyOn is a full-featured edge-to-cloud IoT platform that provides a comprehensive solution for deploying intelligent, AI-driven IoT applications. The platform handles data transformation, transmission, and orchestration, allowing businesses to seamlessly integrate and manage their IoT systems.

How Can One Platform Solve All These Challenges?

1. Lightweight and Efficient Architecture

   InfinyOn is powered by Fluvio, an open-source data streaming platform built in Rust. Its low footprint and memory requirements, combined with high throughput, make it suitable for resource-constrained edge devices. The InfinyOn cluster can be effortlessly deployed on hardware like Raspberry Pi, enabling cost-effective and scalable solutions.

2. Seamless Scalability

   The InfinyOn cluster effortlessly scales for cloud environments, allowing users to orchestrate edge and cloud services from a single interface. This unified approach simplifies management and enhances operational efficiency as businesses grow.

3. Built-in Caching Mechanism

   InfinyOn features built-in caching backed by storage to preserve information during network outages. This means that the edge cluster remains operational for edge producers, regardless of cloud connection status, making it easier for businesses to build and maintain their systems.

4. Reliable Data Mirroring

   InfinyOn includes a mirroring function that leverages a robust replication engine to efficiently replicate data to the cloud. This function is designed to operate reliably in low-bandwidth environments, ensuring minimal impact on network performance while maintaining data integrity.

5. Built-in Runtime for AI Models

   With InfinyOn’s Stateful Data Flows (SDF) engine, powered by WebAssembly, system operators can upload AI models and execute them either on the edge or in the cloud. This flexibility enables organizations to deploy intelligent applications tailored to their specific needs, optimizing performance and enhancing decision-making capabilities.

Conclusion

The InfinyOn architecture simplifies the development and maintenance of end-to-end IoT applications. By addressing critical challenges with a robust and versatile platform, InfinyOn empowers businesses to harness the full potential of their IoT investments, ensuring they stay competitive in an ever-evolving landscape.

To learn more about how InfinyOn can transform your edge data processing power digital twins, predictive maintenance, anomaly detection type use cases contact us for a demo today.

Stay in Touch:

• Join our community on Github Discussions or Discord to share your insights on stateful data flows.
• Subscribe to our YouTube channel for more tutorials and updates.
• Follow us on Twitter for the latest news.
• Connect with us on LinkedIn for professional networking.