Earticle

As autonomous driving and connected car technology advance, various deep learning applications for autonomous vehicles and complex traffic situations are increasing. Autonomous vehicles must collect and process vast amounts of sensor data to support various deep learning applications, but vehicles have limited computing resources to perform complex deep learning operations. Therefore, edge computing is a promising solution to complement the limitations of autonomous vehicles. In this paper, we design edge computing for efficient task processing in an autonomous driving environment using a driving simulator. Also, we propose a task scheduling and offloading method which determines the target server to offload a task according to the characteristics of the task and the computing resources. The effectiveness of the proposed method is verified through experimental evaluation in an autonomous driving environment, supporting multiple deep learning services that we established by using a driving simulator.

디바이스 소셜리티 내에는 서버, 데스크탑, 모바일 단말기와 같은 다양한 성능을 가진 디바이스들이 존재할 수 있 다. 이러한 환경에서 모바일 단말기의 경우 상대적으로 다른 디바이스들에 비해 성능이 낮고 전력 소모에 민감하다. 따라서 모바일 단말기의 제한점을 해결하기 위해선 다른 디바이스의 고성능 자원의 공유가 필요하다. 공유 대상 자 원의 종류 중 GPGPU의 경우 최근에 모바일 단말기에서 지원이 시작되고 있어 성능이 제한적이고 일부 단말기의 경우 지원하지 않는다. 따라서 본 논문에서는 디바이스 소셜리티 내에서 모바일 GPGPU가 장착되지 않은 모바일에 서도 GPGPU 수행을 가능하게 하며, GPGPU 기반 연산의 성능을 향상시킬 수 있는 GPGPU 자원 공유 방식인 GPGPU 오프로딩 프레임워크를 제안하였다. 이 프레임워크는 OpenCL에 기반한 GPGPU 응용을 지원하며, OpenCL 소프트웨어 스택 중 프로그래밍 언어 사이의 이식성이 높고 구현의 복잡도가 낮은 레이어를 대상으로 구 현되었다. 실험결과로 행렬 곱 512×512 워크로드의 경우 오프로딩을 수행한 결과가 로컬 GPGPU를 사용했을 때 의 수행 시간보다 약 2.66배 향상되었다.

In the device sociality, devices which have various performance may exist such as server, desktop or mobile. In this environment, mobile device has poor performance and is sensitive to power consumption. Therefore, it is required to share other device’s high performance resources in order to resolve the limitation of mobile devices. Among the resources, the performance of mobile GPGPU is lower and some mobile devices still do not support GPGPU. In this paper, therefore, we proposed mobile GPGPU offloading framework that enables mobile GPGPU application to run on mobile which has not GPGPU and improves the performance of GPGPU-based application. This framework supports OpenCL-based GPGPU application and is built on the layer which has high portability and low complexity, among OpenCL software stacks. In case of 512×512 matrix multiplication, the result have shown that remote execution performs about 13.5 times better than local execution.

Task offloading in multi-access edge computing (MEC) systems is critical for managing computational tasks in dynamic urban environments. Existing strategies face challenges such as high communication overheads and regional performance deviations including centralized and distributed methods. Clustering approaches have been explored to address these issues, yet they often rely on physical proximity to form clusters, overlooking the variability in task rate distributions across edges. To overcome these limitations, this paper proposes a graph-driven inter-cluster resource distribution (GIRD) clustering scheme that clusters edge nodes based on task request distribution and computing resource status, ensuring similar resource utilization across clusters. Building on this, a proximal policy optimization (PPO)-enabled intra-cluster task offloading algorithm (PITO) is introduced to determine one execution server for task offloading?either an edge server within a cluster or a cloud server?using various network state information. This dynamic decision-making process optimizes a multi-objective function that includes task processing delay, consumed energy, success rate, and cloud cost. Simulation results demonstrate the proposed GIRD-PITO framework achieves superior task success rates, reduced delays, and improved regional performance fairness, making it a promising solution for large-scale MEC systems. 2018 The Korean Institute of Communications and Information Sciences. Publishing Services by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Despite the rapid growth of the Internet industry, the provision of full Internet service to remote regions is still challenging. As a solution, the combination of Low Earth Orbit (LEO) satellite communication and Mobile Edge Computing (MEC) is gaining attention. However, considering the high speed of LEO satellites in network environments remains a significant challenge. To this end, this paper introduces a dynamic computation offloading and resource allocation framework in the LEO satellite MEC architecture. Using Lyapunov optimization, we propose an efficient DCOOL algorithm to minimize average power consumption and propagation delay constrained by queue stability. Finally, comparative analysis and simulations demonstrate the superior performance of DCOOL while achieving lower power consumption and stable workload processing.

In this paper, end devices are considered here as agent, which makes its decisions on whether the network will offload the computation tasks to the edge devices or not. To tackle the resource allocation and task offloading, paper formulated the computation resource allocation problems as a sum cost delay of this framework. An optimal binary computational offloading decision is proposed and then reinforcement learning is introduced to solve the problem. Simulation results demonstrate the effectiveness of this reinforcement learning based scheme to minimize the offloading cost derived as computation cost and delay cost in industrial internet of things scenarios.

Edge computing has emerged as a promising paradigm for addressing the latency, bandwidth, and scalability challenges associated with traditional cloud-centric architectures. Computation offloading, the process of transferring computational tasks from edge devices to more powerful remote servers or cloud infrastructure, plays a crucial role in optimizing performance and resource utilization in edge computing systems. However, traditional computation offloading techniques often face limitations related to latency, network dependency, and scalability. In this survey, we explore the integration of digital twin (DT) technology into edge computing environments to empower intelligent computation offloading decisions. DTs, virtual representations of physical entities or systems that mirror their real-world counterparts, offer opportunities to enhance situational awareness, optimize resource allocation, and enable more informed decision-making at the edge. We provide a comprehensive overview of DTs empowered intelligent computation offloading, covering the fundamentals of DTs, traditional computation offloading techniques, and their limitations in edge computing. Additionally, we discuss how DTs can address these challenges and improve computation offloading strategies, along with practical applications and use cases across various domains. Finally, we identify open research challenges and opportunities for future exploration in this emerging field. Through this survey, we aim to provide researchers, practitioners, and stakeholders with insights into the potential of DTs to revolutionize computation offloading for edge computing and drive innovation in this rapidly evolving area.

This paper investigates QoS-aware computation offloading issues for mobile edge computing in the 6G network. To minimize the end-to-end delay, we harness the Information-Centric Network (ICN) to ensure resource-constrained mobile user offloading computation-sensitive tasks in a distributed manner. Then, a two-stage approach based on a Multi-Agent Reinforcement Learning (MARL) algorithm entwined with optimization-embedding offloading ratio is proposed to enhance server selection for load balancing. Numeral results demonstrate that, with reference to a workshop-scale scenario, the proposed method can achieve outperformed performance in reducing delay and balancing loads on edge servers than the other four baseline schemes.

We address the issues of difference in latency tolerance with heterogeneous IoT users from ground and aviation space by proposing a collaborative satellite terrestrial EDGE computing network. Based on the variability in latency tolerance we propose three offloading schemes under two distinctive scenarios. Optimization in resource sharing while offloading is carried out by adopting DDPG-based actor-critic framework which is suggested as a suitable algorithm by recent studies. We validated the schemes evaluating four performance parameters. Results showed that schemes that tolerate delays between 0.25 to 2.00 s outperformed other schemes in terms of reward, delay and energy consumption.

Cellular networks are facing explosive growth of mobile data traffic due to the proliferation of smart devices and traffic-intensive applications. As a cost-effective solution, delayed Wi-Fi offloading was introduced to shift the delay-tolerant traffic from cellular to Wi-Fi networks by trading additional delays. Our paper studies a multi-flow rate control problem where each flow has different traffic load and a deadline. To maximize user satisfaction defined as offloading efficiency minus disutility caused by deadline violation, we propose a dynamic programming-based rate control algorithm. Moreover, to reduce the computation and memory, we propose a simple threshold-based rate control algorithm.

Beyond 6G services and applications demand high and efficient processing capacity due to the massive connectivity of users equipment (UEs). However, the high computational capability and energy consumption of UEs are limited, which becomes a main challenge to overcome. Multi-access edge computing (MEC) has recently been studied widely as it can potentially assist complex tasks executed at UEs. Furthermore, several techniques have been proposed to optimize task offloading among users. Thus, another challenge in MEC is emerging due to the fact that mobile users do not always have a line-of-sight (LoS) to the base station (BS) due to the blocking object. Therefore, it can affect users data rate and result in incremental energy consumption. This research introduces the concept of reconfigurable intelligence surfaces (RIS) to support multiple-input-single-output (MISO) base stations (BS) in both uplink (UL) and downlink (DL) using BCD algorithms. While previous studies concentrate on enhancing task offloading and neglecting inter-user interference, this study suggests an optimization approach for UL and DL data rates, as well as minimizing task offloading delays. The results indicate that optimizing task placement, phase shift, and precoding can reduce the duration of task offloading.

The mobile device has been widely used in many areas. Task offloading is always used to overcome the limitation of processing ability and energy-supply of the mobile devices in the mobile cloud environment. Most researches suppose we know the detail of task (number of instructions) of every sub-task. But most of time, to get the detail of a task is very difficult. This paper uses the grey number to express that uncertain information and propose method to evaluate the performance of offloading tasks. Simulations show that our method selects the solution with the lowest energy consumption.

Edge computing is a promising technique for processing real time data that is required in various fields such as smart cities, autonomous driving and connected cars. In general, edge computing has limited resources in comparison with the central cloud. It is a greater disadvantage for the embedded systems if the nodes in the edge cluster has no sufficient resources to process the requests. In this paper we propose a service offloading technique which solves the problem of resource inefficiency in Kubernetes based edge environment. The proposed method monitors the capacity of the edge nodes and performs service offloading if necessary. The preliminary experiment is performed to monitor the resource usage by an edge node. In addition, the proposed method aims to reduce the idle time and waiting time of the requests during offloading which inturn reduces the total latency duration.

IoT 센서 네트워크에서는 즉각적인 처리가 어려울 정도의 방대한 양의 데이터가 매 초마다 수집되므로, 상당한 양의 데이터를 미리 선별하고 무가치한 데이터를 신속히 제거하는 오프로딩 기술이 매우 중요하다. 본 연구에서는 개 별 센서로 부터 받은 데이터를 고장 패턴을 기준으로 복합적으로 해석하여, 센서 네트워크의 데이터 부하와 전송 부하 를 효율적으로 제어하는 복합 센서 컨텍스트 기반 다층 모델과 징후 패턴 기반 적응형 오프로딩 제어 기법을 제안한다. 즉, 해당되는 센서 입력들을 비교 가능한 형식으로 스케일링 패턴화 하고, 이를 그리드 패턴 매칭을 수행하여 컨텍스트 매칭 성공 여부를 결정한다. 따라서 실제 고장의 원인일 확률이 높은 센서 정보를 선별하고 전송하여 결과적으로 오프 로딩 성능을 향상시키고, 응답 지연 시간을 크게 줄일 수 있다. 그리고 대량의 데이터 학습이 필요한 인공지능 기법에 비해 제안 기법은 컴퓨팅 리소스가 제한된 에지 게이트웨이에서 가벼운 통계 계산을 사용하여 최적의 센서 조합을 신속하게 추적할 수 있다는 장점이 있다. 또한 빠른 데이터 스크리닝을 위해 적응형 오프로드 모델에서 센서 후보 필터 링을 고안하였다. 테스트 결과는 공개 데이터베이스에서 고장 예측 정확도가 최대 89%까지 증가함을 보였다.

Since vast amounts of data are collected every second, immediate processing is difficult in IoT sensor networks. Thus, the offloading skill quickly eliminating worthless data and selecting significant data is very important. In this work, we propose a composite sensor context-based multiple layer model and a adaptive offloading technique that efficiently controls the data and transmission workload by interpreting data contexts from individual sensors with fault patterns. That is, the corresponding sensor inputs are scaled in a comparable grid format, and pattern matching is performed on the grid to determine whether the context matching is successful. Therefore, the sensor information that is highly likely to be the cause of the actual device failure can be selected and transmitted, resulting in improved offloading performance and reduced response time. Compared to artificial intelligence technologies which need heavy data learning, the proposed scheme has the advantage of being able to quickly track optimal sensor combinations using light statistical calculations on the edge gateways which have limited computing resources. This study devised sensor candidate filtering in adaptive offloading model for fast data screening. The test results show that the fault prediction accuracy increased up to 89% in a public database.

본 논문에서는 GPGPU 가 없거나 저성능 GPGPU를 가진 시스템에서도 고성능 GPGPU 응용을 수행할 수 있도 록, 로컬의 GPGPU응용을 오프로딩 하여 원격지에서 수행 시키는 프레임워크를 제안한다. 이 프레임워크는 응용으 로부터 호출된 API를 OpenCL 라이브러리를 대신하여 새롭게 정의된 원격 전송 라이브러리를 통해 원격지로 보내 고, 원격지에서는 전송 받은 API를 수행한 결과 값을 로컬로 전송한다. 이 결과 값은 로컬측 원격 전송 라이브러리 를 통해 응용에게 다시 전달되므로, 응용은 하위 라이브러리 계층에 투명하게 원격지 수행을 보장 받을 수 있다. 제 안하는 프레임워크 프로토타입은 OpenCL 버전 1.1을 지원하며, 원격 수행 시 기존 응용의 수정을 필요로 하지 않 는다. Matrix Multiplication의 다양한 사이즈를 이용한 실험결과 원격지 수행이 로컬 수행 보다 최대 4.9배 향상 됨을 볼 수 있었으며, LavaMD 실험의 경우 최대 3.2배 정도의 수행 성능 향상을 볼 수 있었음.

In this paper, we propose the framework that enables a local GPGPU application to run in remote site by offloading to support the system which does not have the GPGPU or has low performance GPGPU. This framework sends the invoked APIs from the application to the remote site using the newly defined remote transmission library instead of the original OpenCL library. In the remote site, the execution results of the received APIs are sent to the local machine. Since these results are transferred to the user application through the remote transmission library in the local site, the application is guaranteed the remote execution transparently. The prototype of this framework supports OpenCL version 1.1 and it does not require any modification of the application when it runs in the remote site. The experimental results for remote execution have shown that it performs maximum 4.9 times better than local execution using various Matrix Multiplication size and maximum 3.2 times better than local execution using the LavaMD workload.

차세대 네트워크에서는 제한된 자원을 가진 수많은 장치가 실시간 응용 서비스를 위해 지연에 민감 하고 복잡한 작업을 수행해야 함에 따라 작업 오프로딩과 자원 할당이 점차 중요해지고 있다. 이는 복 잡한 결정 문제로 사전 지식 없이 최적 결정을 선택할 수 있는 강화 학습을 통해 해결할 수 있다. 본 연구는 이를 위한 연합 심층 강화 학습의 연구 동향을 정리하고, 자원 제약과 학습 구조를 중심으로 분석한 후 향후 연구 방향을 제시하고자 한다.

In our research work, we will propose a lightweight energy estimation framework that will enable mobile users to acquire the power profile of desired mobile model from the web to estimate application power consumption while suppressing the need to power log time-stamped execution traces during application execution on mobile phone to construct power models. For the experimentation purpose, we will chose some standard benchmarks and google app store applications (such as ‘whet-stone’ and ‘LINPAC’) to highlight the energy estimation accuracy when considering instruction level power estimation methodology.

오프로딩(offloading)은 컴퓨팅 자원 및 계산 속도의 한계를 극복하기 위해 로컬 컴퓨터에서 수행하는 어플리케이션 의 일부를 컴퓨팅 자원과 처리능력이 우수한 원격지 컴퓨터에 전달하여 처리한 후 결과를 반환받는 방식이다.최근에는 컴퓨 팅 자원과 처리능력에 한계를 갖고 있는 모바일 컴퓨팅 분야에서 처리속도를 높이고 배터리 소모를 줄이기 위해 모바일 게임, 멀티미디어 데이터, 360도 동영상 처리, 인터넷 방송용 이미지 처리 분야에서 응용되고 있다. 본 논문에서는 모바일 장치에서 촬영된 360도 구면 영상을 사용자가 쉽게 영상 내용을 이해할 수 있도록 다양한 평면 영상을 변환하고 내용을 확인할 수 있는 뷰어를 무선 인터넷 환경에서 오프로딩 기술을 적용하여 구현하고 실제 실험 결과를 제시한다. 360도 구면 영상은 인터 페이스를 통해 360도 카메라의 영상 획득 위치에 따라 Double Panorama, Quad, Single Rectabgle, 360 Overview + 3 Rectangle로 평면영상으로 변환이 성공적으로 수행된다. 실험과정에서 100가지 이상의 360도 구면 영상을 아래 인터페이스 를 통해 평면 영상으로 변환을 성공적으로 진행하였다.

Offloading is a method of communicating, processing, and receiving results from some of the applications performed on local computers to overcome the limitations of computing resources and computational speed.Recently, it has been applied in mobile games, multimedia data, 360-degree video processing, and image processing for Internet broadcasting to speed up processing and reduce battery consumption in the mobile computing sector. This paper implements a viewer that enables users to convert various flat-panel images and view contents in a wireless Internet environment and presents actual results of an experiment so that users can easily understand the images. The 360 degree spherical image is successfully converted to a plane image with Double Panorama, Quad, Single Rectangle, 360 Overview + 3 Rectangle depending on the image acquisition position of the 360 degree camera through the interface. During the experiment, more than 100 360 degree spherical images were successfully converted into plane images through the interface below.