[Paper Review] Network Slicing for Ultra-Reliable Low Latency Communication in Industry 4.0 Scenarios

Anders E. Kalør, René Guillaume+3 more|arXiv (Cornell University)|Aug 30, 2017

Software-Defined Networks and 5GComputer Science8 references19 citations

TL;DR

This paper proposes a network slicing framework for Industry 4.0 that decouples deterministic and packet-switched industrial protocols from underlying physical technologies, enabling end-to-end reliability and latency guarantees using network calculus. It demonstrates that overwriting-based slicing achieves high reliability (1−10⁻⁶) and low latency (≤1 ms) under realistic traffic loads, with isolation maintained via priority queuing and traffic shaping.

ABSTRACT

An important novelty of 5G is its role in transforming the industrial production into Industry 4.0. Specifically, Ultra-Reliable Low Latency Communications (URLLC) will, in many cases, enable replacement of cables with wireless connections and bring freedom in designing and operating interconnected machines, robots, and devices. However, not all industrial links will be of URLLC type; e.g. some applications will require high data rates. Furthermore, these industrial networks will be highly heterogeneous, featuring various communication technologies. We consider network slicing as a mechanism to handle the diverse set of requirements to the network. We present methods for slicing deterministic and packet-switched industrial communication protocols at an abstraction level that is decoupled from the specific implementation of the underlying technologies. Finally, we show how network calculus can be used to assess the end-to-end properties of the network slices.

Motivation & Objective

• To address the challenge of supporting diverse, coexisting communication requirements—such as ultra-reliable low-latency control, high-data-rate updates, and massive machine connectivity—in industrial networks.
• To enable network slicing that is independent of specific underlying technologies (e.g., URLLC, TSN, legacy protocols), focusing on abstracted protocol characteristics.
• To provide end-to-end latency and reliability guarantees for industrial network slices using deterministic network calculus (DNC).
• To evaluate trade-offs between utilization, reliability, and isolation in slicing schemes for mixed deterministic and switched traffic.

Proposed method

• The authors model industrial communication protocols at an abstraction level decoupled from physical layer implementations, focusing on cycle time, frame size, and transmission patterns.
• They propose two slicing schemes: overwriting (reusing reserved time slots) and prioritized queuing (using priority queues to isolate traffic).
• End-to-end latency and failure rate bounds are computed using deterministic network calculus (DNC), which accounts for service curves and arrival curves in the network path.
• The approach models both deterministic (cyclic) and packet-switched (e.g., IP-based) traffic, enabling analysis across heterogeneous industrial networks.
• Traffic shaping via token buckets or reserved resource limits is applied to bound the number of low-priority frames and maintain latency constraints.
• A case study on a personalized medicine manufacturing system validates the method with real-world traffic profiles, including control, alarm, and patient data requests.

Experimental results

Research questions

• RQ1 How can network slicing be designed to simultaneously support URLLC, eMBB, and mMTC requirements in a single industrial network?
• RQ2 What are the trade-offs between utilization, reliability, and isolation when slicing deterministic and packet-switched industrial protocols?
• RQ3 Can network calculus provide accurate end-to-end latency and reliability bounds for network slices spanning heterogeneous communication technologies?
• RQ4 How does prioritized queuing affect latency isolation between high- and low-priority traffic in a factory network?
• RQ5 What is the impact of alarm traffic burstiness on end-to-end latency and reliability in a sliced industrial network?

Key findings

• The overwriting slicing scheme achieves a failure rate of 1−10⁻⁶ for control traffic, sufficient for 60-second inter-arrival times, with 100% resource utilization.
• Alarm traffic latency approaches 1 ms when arrival rates are low, increasing with higher alarm frequency due to serialization delays.
• Patient information request latency increases with alarm traffic load, showing a steeper slope than alarm latency due to conservative affine bounds in DNC.
• Prioritized queuing introduces limited isolation, as high-priority traffic increases queuing delay for lower-priority flows, especially when multiple priority levels are present.
• The maximum number of alarm frames per cycle is bounded by reserved control frame capacity (e.g., 1 or 4 frames), which can be used to enforce traffic shaping and maintain latency bounds.
• Network calculus enables accurate end-to-end analysis of hybrid deterministic and packet-switched networks, supporting design of reliable, low-latency industrial slices.