Determinism at Scale: Audi’s KPIs and Technical Limits of Virtualized Control

KEYWORDS: Audi, Edge, Cloud, vPLC, Innovation, Scaling, PRP

Overview

What motivated Audi to redesign its traditional shop floor architecture? The automotive industry is changing rapidly in the face of new cost structures driven by the electric vehicle transition and fierce competition from China. In Audi’s view, the traditional model of delivering production equipment floor is outdated, and the future is software-driven production.

Audi calls it “IT-based factory automation,” a concept that is turning IT/OT convergence theory into practice at the premium automaker. To achieve this, Audi kicked off an initiative back in 2018 called Edge Cloud 4 Production (EC4P). The goal is to replace thousands of shop floor industrial PCs with a much more efficient and flexible IT-like architecture of scalable local servers, effectively uniting the cloud and the edge on the shop floor.

This ARC View is part of a series of reports that dig deeper into the culture and technical aspects of EC4P.

Key Takeaways

• Scaling vPLCs is as much organizational as it is technical, requiring cultural change, new operational models, and close collaboration between IT, OT, and vendors. 
• Deterministic real‑time performance is achievable but requires careful tuning. 
• Networking is one of the primary scaling bottlenecks. 
• Current standards and products limit large‑scale deployments. 
• Production rollout must be slow, measured, and repeatable.

Scaling vPLCs on the Shop Floor

The transition to virtualized and software‑defined automation solutions represents a significant cultural, technical, and operational shift for shop floor environments. While these approaches promise flexibility, scalability, and long‑term efficiency gains, they also introduce constraints that differ fundamentally from traditional operational technology (OT) models. This ARC View describes the challenges encountered when scaling such solutions from laboratory environments to the shop floor, with a particular focus on real‑time performance, networking constraints, and ecosystem development.

Scaling Challenges

Scaling a new solution on the shop floor as disruptive as EC4P involves the introduction of new technologies that may not align cleanly with existing operations models. Many of the tools and platforms involved were originally designed for IT environments and are not optimized for OT requirements such as deterministic behavior, strict cycle times, and fieldbus‑based communication such as PROFINET. As a result, there has been limited product development specifically targeting these use cases to date.

To bridge this gap, it is necessary to combine technological scaling with a broader cultural shift. Close collaboration with partners is essential so that they understand the specific requirements of industrial automation environments. Vendor engagement and support play a critical role, particularly where existing products must be adapted or extended to meet real‑time and safety‑critical constraints.

Virtualization and Real‑Time Performance

Experience from industries such as finance and telecommunications shows that demanding real time and availability requirements can be met through careful configuration of virtualization platforms. A central challenge in EC4P is bringing multiple virtual PLCs onto the same host while maintaining deterministic performance. The Audi team and its key partners have transferred these learnings to industrial automation, using the following mechanisms, known as high-latency sensitivity configuration:

• Exclusive CPU Access: Assigns vPLC virtual CPUs to physical cores, stopping hypervisor interruptions so the guest OS runs almost natively.
• Immediate Network Processing: Disables network interrupt coalescing, meaning network packets are processed instantly instead of being grouped together to save CPU cycles.
• Strict NUMA: This applies to servers with more than one socket by forcing the VM's memory and CPU to stay on the same physical node, preventing latency spikes caused by fetching remote memory.

These optimizations require 100 percent CPU and 100 percent memory reservations for the vPLC.

For seamless integration and redundancy of virtualized OT control, PROFINET (industrial Ethernet) and the Parallel Redundancy Protocol (PRP) for seamless failover, protocol support was extended into the virtualization stack. Bundled in the Industrial vSwitch, this allows virtualized OT control applications to benefit from platform technologies such as:

• VMXNET3 para-virtualized network hardware
• Enhanced data path (EDP) dedicated with DPDK
• End-to-end strict latency and jitter enforcement

Because real‑time and safety requirements are critical and some of that functionality is hardware-related, a thorough hardware selection process is required. The server hardware must be adequate to run real-time applications, and the network cards must support EDP.

End‑to‑End Connectivity and Networking Constraints

Scaling vPLCs is fundamentally constrained by end‑to‑end connectivity. Data paths span multiple layers, increasing sensitivity to latency, jitter, packet loss, and a chance frame reordering or similar. For PROFINET, support for 1ms cycle times is critical and consecutive frame loss can immediately halt production. Redundancy mechanisms such as PRP impose additional requirements, including larger node tables and VLAN‑aware control planes. These challenges were addressed through the IVS and properly configured commodity hardware capable of deterministic, low‑latency Layer‑2 networking, all of which require extensive validation under realistic operating conditions. Ultimately, success no longer depends on proprietary hardware, but on correctly engineered and configured hardware‑software combinations.

Building an Ecosystem and Addressing Standards

Scaling beyond isolated implementations required the development of a broader ecosystem, meaning working within the constraints of existing IT and OT standards while acknowledging their limitations. For widespread industry adoption, even seemingly minor technical issues, sometimes affecting only a small portion of the overall system, can become critical blockers.

In these cases, timely and structured feedback to standards bodies is essential. One example is the need for VLAN‑aware PRP control planes, which may require extensions to the IEC 62439‑3 standard that defines PRP. Close collaboration with vendors and standardization organizations is therefore necessary to ensure that such requirements are addressed in a way that enables long‑term scalability and interoperability. It was critical to build a testing environment where all technologies could be tested together for application, scale and failover testing to provide confidence it would work in production.

Capacity Limits and KPIs

Several quantitative limits currently define the scalability of virtualized PLC solutions:

• The number of vPLCs that can run on a single host is a key factor when laying out a new architecture. The current practical limit is 12 per server, driven by the number of VDANs available on a server.
• In a representative production scenario with 17 PLCs on a line, at least two servers are required, with a third server recommended as a spare.
• Safety couplers influence the physical layout and geographical distribution of equipment and safety PLCs.
• Geographical placement of data centers also plays a role. Server locations up to 10 miles away are feasible, but this adds latency of about 5 μs per kilometer. While this is generally acceptable, it requires a robust networking infrastructure, often involving leased lines, to ensure consistent performance.

From Laboratory to Production

The transition from laboratory environments to production follows a cautious, phased approach. Initial deployments in pilot labs provide valuable real‑world experience, but these environments do not yet represent high‑volume production. Trust in the technology builds gradually through successful operation, but it can be lost quickly if issues arise.

Because automotive production systems are deeply integrated, small technical problems can have widespread impacts. As a result, large, rapid changes are difficult to implement. Each new site requires repetition of the validation and rollout process, as stakeholders need to see the solution working in their own context before accepting it.

Technical Scaling Constraints

From a technical perspective, scaling is constrained by several factors:

• Vendor and product limitations, such as early vPLC solutions lacking safety support or having restrictions on the number of PROFINET IOs.
• Hardware constraints, including the number of available CPU cores.
• Networking constraints, such as PRP node table sizes and the complexity of tracking duplicated traffic.
• Gaps in current industry standards that do not fully address these virtualized, redundant architectures.

Overcoming these limitations requires coordinated efforts across vendors and the broader industry to enable sustainable, large‑scale adoption.

Conclusion

Scaling virtualized PLC solutions to the shop floor is not purely a technical exercise. It requires a combination of architectural rigor, operational adaptation, ecosystem development, and cultural change. While pilot implementations demonstrate technical feasibility, broad adoption depends on resolving standards gaps, vendor limitations, and operational concerns. With the EC4P project, Audi has taken a slow, methodical approach, grounded in measurement, KPIs, and real‑world validation, to build long‑term trust and success.
 

ARC Advisory Group clients can view the complete report at the ARC Client Portal.

Contact Us if you would like to speak with the author.

Obtain more ARC In-depth Research Market Analysis.