Containerized PLCs and edge computing are changing the approach to industrial control. An analysis of how Linux and containers ensure determinism under heavy load.
The problem arises when classical industrial architecture faces the demands for connectivity and real-time data processing. Traditional PLC controllers operate in closed, proprietary environments where determinism is achieved through isolation. This simplifies predictability but complicates scaling, updating, and integration with modern systems. This is particularly evident when trying to combine control loops and data-heavy workloads: competition for CPU and cache leads to jitter and violations of timing guarantees (latency).
The key question is whether control can be transferred to a software-defined environment without losing determinism. A solution is proposed based on a stack using Red Hat Enterprise Linux with a real-time kernel and Red Hat Device Edge in combination with Intel hardware. This approach bets on unifying the environment: a single platform for the data center and edge. The compromise here is clear—an open system provides flexibility and automation but raises concerns about the stability of timing characteristics (screw-to-screw time). This parameter, from the physical signal to the mechanical response, is critical for the industry.
The hypothesis was tested around a virtual PLC running on RHEL. The scenario was intentionally complicated: resource-intensive applications that actively consumed CPU and cache were added to the system. This models real workload consolidation, where analytics and control operate side by side. Without additional isolation mechanisms, variability increased. However, the use of Intel Cache Allocation Technology (CAT) changed the system’s behavior. CAT restricts process access to the cache, reducing mutual influence. As a result, even with CPU loads at 80%, the worst-case jitter for interrupt requests remained below 30 microseconds.
An unexpected result came from the level of container isolation. Running control logic inside Podman showed more stable timing characteristics compared to the bare-metal scenario with active CAT. This indicates that containerization can not only maintain determinism but also stabilize it through more predictable resource management. The advantages of the container model are preserved: portability, standardized deployment, and lifecycle management.
The practical implication is a shift in the operational model. Instead of a rigid tie to specialized hardware, there is an opportunity to use a unified stack for edge and core. This simplifies patch management, security policies, and updates. The infrastructure becomes closer to cloud-native principles, even at the factory floor level. Additionally, there is the possibility of processing data directly at the source (edge computing), which reduces dependence on the network and eliminates delays associated with transferring data to the cloud or a local data center.
The results show that the technical barrier to using Linux in industrial control has been significantly lowered. Determinism is maintained even under heavy load and containerization. However, quantitative metrics are limited to the data presented: variability of less than 30 microseconds was recorded, but broader comparisons with alternative architectures are not provided. Nevertheless, the direction is clear—the boundary between IT and OT is gradually blurring.
This approach appears to be a pragmatic evolution. It does not negate the requirements for real-time systems but offers tools to comply with them in a more flexible architecture. The main trade-off shifts from “hardware versus flexibility” to “resource control versus platform universality.” And judging by the tests, this balance is becoming manageable.