Experimental Assessment of Drone Computing Offloading in Real 5G Operator Networks

Core Concepts

The core message of this article is that the feasibility of offloading computationally intensive drone applications to 5G edge computing platforms depends on the specific network architecture and configuration, and that a private carrier-grade 5G standalone network with co-located edge resources can provide significantly better performance compared to a commercial 5G non-standalone network with semi-centralized edge resources.

Abstract

The article presents an experimental assessment of offloading a drone "follow-me" application, which involves processing video from the drone camera to track a person and send positioning commands back to the drone, to real 5G operator networks and edge computing platforms. The key highlights and insights are: The authors implemented a "follow-me" drone application as a representative use case for computationally intensive drone applications that could benefit from offloading. They evaluated the application's performance in two 5G network setups - a commercial non-standalone (NSA) 5G network and a private carrier-grade standalone (SA) 5G network. In each setup, they assessed offloading to both a multi-access edge computing (MEC) platform and a commercial cloud platform. The results show that the private 5G SA network with a co-located MEC platform provided the best performance, with a median total service delay of only 57.052 ms. This was significantly better than the commercial 5G NSA network with a semi-centralized MEC platform, which had a median delay of 68.414 ms, and the cloud offloading scenarios, which had median delays of 105.632 ms (private 5G) and 140.117 ms (commercial 5G). The performance differences were mainly due to the communication latency, which was much lower in the private 5G SA network setup compared to the commercial 5G NSA network. The processing times were similar across the different platforms. The authors also evaluated onboard computing on the drone and found that it could not meet the latency requirements of the "follow-me" application, highlighting the need for offloading in such computationally intensive drone applications. The results provide guidance on the network configurations that can feasibly support the "follow-me" application use case, depending on the mobility of the end user. They also demonstrate the advantages of MEC over a commercial cloud platform for such latency-sensitive applications.

Stats

The image transmission delay had a median value of 44.163 ms and a standard deviation of 5.266 ms in the commercial 5G network with cloud offloading. The image transmission delay had a median value of 25.594 ms and a standard deviation of 2.379 ms in the private 5G network with cloud offloading. The command transmission delay had a median value of 39.067 ms and a standard deviation of 1.385 ms in the commercial 5G network with cloud offloading. The command transmission delay had a median value of 22.259 ms and a standard deviation of 0.993 ms in the private 5G network with cloud offloading. The image processing delay had a median value of 55.696 ms and a standard deviation of 4.362 ms in the commercial 5G network with cloud offloading. The image processing delay had a median value of 57.091 ms and a standard deviation of 5.152 ms in the private 5G network with cloud offloading. The image processing delay had a median value of 34.942 ms in the private 5G network with MEC offloading.

Quotes

"The core message of this article is that the feasibility of offloading computationally intensive drone applications to 5G edge computing platforms depends on the specific network architecture and configuration, and that a private carrier-grade 5G standalone network with co-located edge resources can provide significantly better performance compared to a commercial 5G non-standalone network with semi-centralized edge resources." "The results show that the private 5G SA network with a co-located MEC platform provided the best performance, with a median total service delay of only 57.052 ms. This was significantly better than the commercial 5G NSA network with a semi-centralized MEC platform, which had a median delay of 68.414 ms, and the cloud offloading scenarios, which had median delays of 105.632 ms (private 5G) and 140.117 ms (commercial 5G)."

Key Insights Distilled From

Is the edge really necessary for drone computing offloading? An experimental assessment in carrier-grade 5G operator networks

by Davi... at arxiv.org 04-01-2024

https://arxiv.org/pdf/2403.19729.pdf

Deeper Inquiries

How would the results change if the drone was moving at higher speeds, such as 50 km/h or 100 km/h, rather than the speeds considered in the article?

If the drone was moving at higher speeds, such as 50 km/h or 100 km/h, the results of the study would likely show increased challenges in maintaining the required service delay for the "follow-me" application. At higher speeds, the processing and communication requirements would need to be adjusted to account for the faster movement of the drone and the user being tracked. This would impact the feasibility of offloading computation to the edge or cloud platforms, as the system would need to respond more quickly to changes in the user's position. The image processing tasks would need to be optimized for faster processing times to keep up with the movement of the drone and the user. Additionally, the communication delays between the drone and the edge or cloud platforms would become more critical at higher speeds, as the system would need to transmit and receive data more frequently to maintain real-time tracking.

What are the potential trade-offs between the performance benefits of a private 5G SA network with co-located MEC and the additional cost and complexity of deploying such a network infrastructure?

The potential trade-offs between the performance benefits of a private 5G SA network with co-located MEC and the additional cost and complexity of deploying such a network infrastructure include: Performance Benefits: Lower latency and higher throughput due to the proximity of the MEC resources to the end users. Improved reliability and security as the network is dedicated and isolated from external users. Enhanced control over network resources and service quality. Cost and Complexity: Higher initial deployment costs for setting up a private 5G SA network with dedicated MEC resources. Ongoing maintenance and operational costs for managing and upgrading the network infrastructure. Increased complexity in network management and coordination compared to utilizing existing commercial networks. The decision to invest in a private 5G SA network with co-located MEC would depend on the specific requirements of the drone applications and the organization's budget and long-term strategic goals. Organizations would need to weigh the performance benefits against the associated costs and complexities to determine if the investment is justified.

How could the authors extend this work to explore the feasibility of offloading other types of computationally intensive drone applications, such as 3D mapping or object detection, and how would the performance requirements and constraints differ from the "follow-me" use case?

To explore the feasibility of offloading other types of computationally intensive drone applications, such as 3D mapping or object detection, the authors could consider the following steps: Application Analysis: Identify the specific requirements and constraints of the new drone applications, such as the processing power needed, data transmission rates, and real-time processing capabilities. Experimental Setup: Modify the existing testbed to accommodate the new applications, including adjusting the image processing algorithms, communication protocols, and data transmission rates. Performance Evaluation: Measure the service delay, processing times, and communication latency for the new applications in different network setups (private 5G SA, commercial 5G NSA, etc.). Analyze the impact of the application requirements on the feasibility of offloading computation to edge or cloud platforms. Comparison: Compare the performance requirements and constraints of the new applications with the "follow-me" use case to understand the differences in processing demands, communication needs, and real-time processing capabilities. Evaluate the trade-offs between performance benefits and costs for each type of application to determine the most suitable network setup for each scenario.