Reducing end-to-end delay in safety-critical systems, such as controlling remote robots with motion controllers in industrial sites, is a major challenge. 5G mobile communication has greatly contributed to reducing tr...
详细信息
ISBN:
(纸本)9781665462686
Reducing end-to-end delay in safety-critical systems, such as controlling remote robots with motion controllers in industrial sites, is a major challenge. 5G mobile communication has greatly contributed to reducing transmission delay, but realtimeness in an edge server system responsible for packet processing and delivery in the middle of an industrial site can contribute even more. Although the Linux community released a kernel patch to enhance the real-time tasks, it occasionally violates the demand of bounded (worst case) latency. In this paper, we propose a real-time scheduling approach in the edge computing server to handle a latency violation and applies it to the two-way motion control system. As a result, our proposed approach shows that data packet processing is more effective than other comparison groups.
GPU-aware collective communication has become a major bottle-neck for modern computing platforms as GPU computing power rapidly rises. A traditional approach is to directly integrate lossy compression into GPU-aware c...
详细信息
ISBN:
(纸本)9798400706103
GPU-aware collective communication has become a major bottle-neck for modern computing platforms as GPU computing power rapidly rises. A traditional approach is to directly integrate lossy compression into GPU-aware collectives, which can lead to serious performance issues such as underutilized GPU devices and uncontrolled data distortion. In order to address these issues, in this paper, we propose gZCCL, a first-ever general framework that designs and optimizes GPU-aware, compression-enabled collectives with an accuracy-aware design to control error propagation. To validate our framework, we evaluate the performance on up to 512 NVIDIA A100 GPUs with real-world applications and datasets. Experimental results demonstrate that our gZCCL-accelerated collectives, including both collective computation (Allreduce) and collective data movement (Scatter), can outperform NCCL as well as Cray MPI by up to 4.5x and 28.7x, respectively. Furthermore, our accuracy evaluation with an image-stacking application confirms the high reconstructed data quality of our accuracy-aware framework.
Fifth generation wireless (5G) technology includes a range of service categories, notably ultra-reliable low-latency communications (uRLLCs). uRLLCs use cases are distinguished by their exceptional reliability in deli...
详细信息
ISBN:
(纸本)9798350387919;9798350387902
Fifth generation wireless (5G) technology includes a range of service categories, notably ultra-reliable low-latency communications (uRLLCs). uRLLCs use cases are distinguished by their exceptional reliability in delivering packets quickly. However, fulfilling the stringent low latency demands of uRLLC applications poses a significant challenge for 5G networks. Traditional central cloud-based architectures struggle to achieve these tight requirements, necessitating creative latency optimisation methods. Therefore, Edge computing emerges as a viable paradigm, with the potential to bring computational resources closer to end users and applications to achieve end-to-end latency of around 10 milliseconds (ms). In this paper, we propose an efficient architecture for minimising latency in 5G networks by moving the user plane function (UPF) to the network edge closer to the users, leveraging the control and user plane separation (CUPS) strategy. In addition, this paper further proposes the software-defined networking (SDN) based backhaul. Leveraging SDN in the backhaul network of 5G deployments allows operators to create dynamic, scalable, and efficient networks capable of serving a diverse variety of services and applications with varying performance needs. The experiment results from the 3rd generation partnership project (3GPP) compliant 5G testbed prove that the proposed architecture optimises the latency by 60%. Average throughput is also improved by approximately 40%.
The health and safety of miners are seriously at stake since coal mining is a difficult and dangerous job. Innovative solutions have been created to reduce these dangers and enhance the efficiency and safety of coalmi...
详细信息
With the continuous expansion of optical fiber communication network scale, rapid growth of capacity and increasingly rich services, it has become a major challenge for the future optical communication network to seek...
详细信息
American Sign Language is a hand sign language that allows two people to communicate completely by moving their hands together. This sign language is useful for mute and deaf individuals to communicate about their sit...
详细信息
The thermal reliability of servo control drivers is an important link in the thermal reliability of aerospace aerodynamic control electromechanical servo systems, and the modeling and simulation analysis of their ther...
详细信息
As Video Conferencing plays an increasingly critical role in many business environments, there is a need to ensure highly reliable operation of the conferencing infrastructure. There are several Video Conferencing App...
详细信息
Based on virtual reality, this paper proposes a robot step length automation planning algorithm based on a linear inverted pendulum model. The algorithm first obtains the position of the center of mass when the suppor...
详细信息
The integration of the Internet of Things (IoT) in sericulture, or silk farming, has the potential to significantly enhance the efficiency and output of silk production. Traditional methods, such as manual monitoring ...
详细信息
暂无评论