• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.7
• /
• pp.1749-1773
• /
• 2024

Recent advances in deep neural networks (DNNs) have greatly improved the accuracy and universality of various intelligent applications, at the expense of increasing model size and computational demand. Since the resources of end devices are often too limited to deploy a complete DNN model, offloading DNN inference tasks to cloud servers is a common approach to meet this gap. However, due to the limited bandwidth of WAN and the long distance between end devices and cloud servers, this approach may lead to significant data transmission latency. Therefore, device-edge collaborative inference has emerged as a promising paradigm to accelerate the execution of DNN inference tasks where DNN models are partitioned to be sequentially executed in both end devices and edge servers. Nevertheless, collaborative inference in heterogeneous edge environments with multiple edge servers, end devices and DNN tasks has been overlooked in previous research. To fill this gap, we investigate the optimization problem of collaborative inference in a heterogeneous system and propose a scheme CIS, i.e., collaborative inference scheme, which jointly combines DNN partition, task offloading and scheduling to reduce the average weighted inference latency. CIS decomposes the problem into three parts to achieve the optimal average weighted inference latency. In addition, we build a prototype that implements CIS and conducts extensive experiments to demonstrate the scheme's effectiveness and efficiency. Experiments show that CIS reduces 29% to 71% on the average weighted inference latency compared to the other four existing schemes.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.10a
• /
• pp.689-692
• /
• 2021

This paper is a prior study to improve the efficiency of offloading based on mobile agents to optimize allocation of computing resources and reduce latency that support user proximity of application services in a Fog/Edge Computing (FEC) environment. We propose an algorithm that effectively reduces the execution time and the amount of memory required when extracting optimal moving patterns from the vast set of spatio-temporal movement history data of moving objects. The proposed algorithm can be useful for the distribution and deployment of computing resources for computation offloading in future FEC environments through frequency-based optimal path extraction.

• Ocean Systems Engineering
• /
• v.7 no.1
• /
• pp.53-74
• /
• 2017

Ship shaped FPSO (Floating Production, Storage and Offloading) units are the most commonly used floating production units to extract hydrocarbons from reservoirs under the seabed. These structures are usually much larger than general cargo ships and have their natural frequency outside the wave frequency range. This results in the response to first order wave forces acting on the hull to be negligible. However, second order difference frequency forces start to significantly impact the motions of the structure. When the difference frequency between wave components matches the roll natural frequency, the structure experiences a significant roll motion which is also termed as second order roll. This paper describes the theory and numerical implementation behind the calculation of second order forces and motions of any general floating structure subjected to waves. The numerical implementation is validated in zero speed case against the commercial code OrcaFlex. The paper also describes in detail the popular approximations used to simplify the computation of second order forces and provides a discussion on the limitations of each approximation.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2021.05a
• /
• pp.100-103
• /
• 2021

IoT 와 인공지능을 접하려는 시도는 최근 들어서 많은 발전을 보이고 있다. 본 논문은 컴퓨팅 파워가 제한되는 작은 디바이스 IoT 의 한계를 극복하기 위하여 ROS 를 이용하여 복잡한 연산을 무선 통신으로 오프로딩하는 기법을 제안한다. 제안된 자율주행카드 시스템은 카트 이용 고객 개개인을 검출하고 추적하되 컴퓨터 비전 알고리즘과 LiDAR 센서를 이용하며, 음성인식 알고리즘을 적용하여 기계와 인간의 감성공학적 소통이 가능한 융합형 자율주행카트를 구현한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.05a
• /
• pp.52-55
• /
• 2022

5G 시대에 스마트 모바일 기기가 기하급수적으로 증가하면서 멀티 액세스 엣지 컴퓨팅(MEC)이 유망한 기술로 부상했다. 낮은 지연시간 안에 계산 집약적인 서비스를 제공하기 위해 MEC 서버로 오프로딩하는 특히, 태스크 도착률과 무선 채널의 상태가 확률적인 MEC 시스템 환경에서의 오프로딩 연구가 주목받고 있다. 본 논문에서는 차량의 전력과 지연시간을 최소화하기 위해 로컬 실행을 위한 연산 자원과 오프로딩을 위한 전송 전력을 할당하는 심층 강화학습 기반의 오프로딩 기법을 제안하였다. Deep Deterministic Policy Gradient (DDPG) 기반 기법과 Deep Q-network (DQN) 기반 기법을 차량의 전력 소비량과 큐잉 지연시간 측면에서 성능을 비교 분석하였다.

• Proceedings of the IEEK Conference
• /
• 2005.11a
• /
• pp.1035-1038
• /
• 2005

This paper describes the designs and the implementations of two H/W IPSec Systems, look-aside and inline, on TOE (Transport Offloading Engine). These systems aim for guaranteeing the security of datagram networks while preserving the bandwidth of gigabit networks. The TOE offloads a host CPU from network burdens, so that it makes the gigabit wire speed possible, and then deeper level security architecture of the IPSec guarantees the security of gigabit service network dominated by datagram packets. The focus of this paper is to minimize the TOE's performance degradation caused by the computation-oriented IPSec. The look-aside IPSec system provides a significant improvement in the CPU offload of the IPSec cryptography loads. However, the inline system completely offloads the host CPU from whole IPSec loads, providing significant additional cost saving compared to the look-aside system. In this paper, the implementations of TOE cards including commercial IPSec processors are presented. As the result of performance evaluation with the protocol analyzer, we can get the fact that the inline IPSec system is 8 times faster than the S/W system and 2 times faster than the look-aside system.

• Journal of Navigation and Port Research
• /
• v.46 no.6
• /
• pp.491-500
• /
• 2022

It has been noted that an accurate estimation of wind loads on offshore structures such as an FLNG (Liquefied Natural Gas Floating P roduction Storage Offloading Units, LNG FPSOs) with a large topside plays an important role in the safety design of hull and mooring system. Therefore, the present study aims to develop a computational model for estimating the wind load acting on an FLNG. In particular, it is the sequel to the previous research by the author. The numerical computation model in the present study was modified based on the previous research. Numerical analysis for estimating wind loads was performed in two conditions for an interval of wind direction (α), 15° over the range of 0° to 360°. One condition is uniform wind speed and the other is the NPD model reflecting the wind speed profile. At first, the effect of sand-grain roughness on the speed profile of the NPD model was studied. Based on the developed NPD model, mesh convergence tests were carried out for 3 wind headings, i.e. head, quartering, and beam. Finally, wind loads on 6-degrees of freedom were numerically estimated and compared by two boundary conditions, uniform speed, and the NPD model. In the present study, a commercial RANS-based viscous solver, STAR-CCM+ (ver. 17.02) was adopted. In summary, wind loads in surge and yaw from the wind speed profile boundary condition were increased by 20.35% and 34.27% at most. Particularly, the interval mean of sway (45° < α <135°, 225° < α < 315°) and roll (60° < α < 135°, 225° < α < 270°) increased by 15.60% and 10.89% against the uniform wind speed (10m/s) boundary condition.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2015.10a
• /
• pp.232-235
• /
• 2015

In recent years, mobile devices are equipped with functionalities comparable to those computers. However, mobile devices have limited resources due to constraints, such as low processing power, limited memory, unpredictable connectivity, and limited battery life. To enhance the capacity of mobile devices, an interesting idea is to use cloud computing and virtualization techniques to shift the workload from mobile devices to a computational infrastructure. Those techniques consist of migrating resource-intensive computations from a mobile device to the resource-rich cloud, or server (called nearby infrastructure). In order to achieve their goals, researchers designed mobile cloud applications models (examples: CloneCloud, Cloudlet, and Weblet). In this paper, we want to highlight on cloudlet architecture (nearby infrastructure with mobile device), its methodology and discuss about the impact of distance between cloudlet and mobile device in our work design.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.19 no.11
• /
• pp.2671-2676
• /
• 2015

Nowadays, Mobile devices are now capable of supporting a wide range of applications. Unfortunately, some of applications demand an ever increasing computational power and mobile devices have limited resources due to their constraints, such as low processing power, limited memory, unpredictable connectivity, and limited battery life. To deal with mobile devices' constraints, researchers envision extending cloud computing services to mobile devices using virtualization techniques to shift the workload from mobile devices to a powerful computational infrastructure. Those techniques consist of migrating resource-intensive computations from a mobile device to the resource-rich cloud, or server (called nearby infrastructure). In this paper, we want to highlight on cloudlet architecture (nearby infrastructure with mobile devices), its functioning and in our future work, analyze effects of distance between cloudlet and mobile devices.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.5
• /
• pp.786-792
• /
• 2022

In the cloud computing environment, there has been a lot of research into the Fog/Edge Computing (FEC) paradigm for securing user proximity of application services and computation offloading to alleviate service delay difficulties. The method of predicting dynamic location change patterns of edge devices (moving objects) requesting application services is critical in this FEC environment for efficient computing resource distribution and deployment. This paper proposes an optimal moving pattern extraction algorithm in which variable weights (distance, time, congestion) are applied to selected paths in addition to a support factor threshold for frequency patterns (moving objects) of edge devices. The proposed algorithm is compared to the OPE_freq [8] algorithm, which just applies frequency, as well as the A^* and Dijkstra algorithms, and it can be shown that the execution time and number of nodes accessed are reduced, and a more accurate path is extracted through experiments.