• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제18권7호
• /
• 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.

• Journal of Communications and Networks
• /
• 제16권5호
• /
• pp.548-558
• /
• 2014

Load distribution is vital to the performance of multipath transport. The task becomes more challenging in real-time multimedia applications (RTMA), which impose stringent delay requirements. Two key issues to be addressed are: 1) How to minimize end-to-end delay and 2) how to alleviate packet reordering that incurs additional recovery time at the receiver. In this paper, we propose sub-packet based multipath load distribution (SPMLD), a new model that splits traffic at the granularity of sub-packet. Our SPMLD model aims to minimize total packet delay by effectively aggregating multiple parallel paths as a single virtual path. First, we formulate the packet splitting over multiple paths as a constrained optimization problem and derive its solution based on progressive approximation method. Second, in the solution, we analyze queuing delay by introducing D/M/1 model and obtain the expression of dynamic packet splitting ratio for each path. Third, in order to describe SPMLD's scheduling policy, we propose two distributed algorithms respectively implemented in the source and destination nodes. We evaluate the performance of SPMLD through extensive simulations in QualNet using real-time H.264 video streaming. Experimental results demonstrate that: SPMLD outperforms previous flow and packet based load distribution models in terms of video peak signal-to-noise ratio, total packet delay, end-to-end delay, and risk of packet reordering. Besides, SPMLD's extra overhead is tiny compared to the input video streaming.

• 대한안전경영과학회지
• /
• 제9권3호
• /
• pp.119-125
• /
• 2007

As competition in manufacturing enterprise is contested, the scope of safely production planning, manufacturing enterprise should ensure, has been reduced. The more upstream of SCM, the more reduction of scope of production planning. As a result, order fluctuation is more sharply contested. Through improving the logistics network, it is best way that the end user's demand information is conveyed to upstream of SCM, but it is difficult in fact. In this paper, it mention the way of robustic adjustment, in the suppliers' point of view, the end user's demand information is dammed up, as they postpone responding the customer's order as a possible. And it will show the result of appling the way, as a case study.

• 한국지능정보시스템학회:학술대회논문집
• /
• 한국지능정보시스템학회 2002년도 춘계학술대회 논문집
• /
• pp.429-434
• /
• 2002

For companies assembling end products from sub assemblies or components, MRP (Material Requirement Planning) logic is frequently used to synchronize and pace the production activities for the required parts. However, in MRP, the planning of operational-level activities is left to short term scheduling. So, we need a good scheduling algorithm to generate feasible schedules taking into account shop floor characteristics and multi-level job structures used in MRP. In this paper, we present a GA (Genetic Algorithm) solution for this complex scheduling problem based on a new gene to reflect the machine assignment, operation sequences and the levels of the operations relative to final operation. The relative operation level is the control parameter that paces the completion timing of the components belonging to the same branch in the multi-level job hierarchy. In order to revise the fixed relative level which solutions are confined to, we apply large step transition in the first step and GA in the second step. We compare the genetic algorithm and 2-phase optimization with several dispatching rules in terms of tardiness for about forty modified standard job-shop problem instances.

• 한국철도학회:학술대회논문집
• /
• 한국철도학회 1998년도 창립기념 춘계학술대회 논문집
• /
• pp.23-30
• /
• 1998

The construction of the transportation history database system is to serve the scheduling and seat inventory controling. Recently, lots of countries have been faced with the advance era because of the new railway transportation system, like the high speed railway and/or magnetic levitation vehicle system. This can be reasonably translated as those of operators are willing to provide the more various and high quality schedule to the customer. Those operators' these ideas make possible to forecast that scheduling process is going to be complicated more and more The seat inventory control, so to speak Yield Management System(YMS), goes a long way to improve the total passenger revenue at the railway business. The YMS forecasts the number of the last reservation value(DCP# END) and recommends the optimal values on the seat sales. The history database system contains infra-data(ie, train, seat, sales) that will be the foundation of scheduling and seat inventory control application programs. The development of the application programs are reserved to the next step. The database system is installed on the pc platform(IBM compatible), using the DB2(RDBMS). And at next step, the platform and DBMS will be considered whether they can meet the users' requirement or not.

• 제어로봇시스템학회논문지
• /
• 제7권7호
• /
• pp.559-566
• /
• 2001

The sway control problem of the pendulum motion of a container hanging on the trolly, which transports containers from a container ship to trucks, is considered in the paper. In the container crane control problem, suppressing the residual swing motion of the container at the end of acceleration, deceleration or the case of that the unexpected disturbance input exists is main issue. For this problem, in general, the trolley motion control strategy is introduced and applied. In this paper, we introduce and synthesize a new type of swing motion control system in which a small auxiliary mass is installed on the spreader. The actuator reacting against the auxiliary mall applies inertial control forces to the container to reduce the swing motion in the desired manner. In this paper, we apply the $H^{\infty}$ based gain-scheduling control technique to the anti-swing motion control system design problem of the controlled plant. In this control system, the controller dynamics are adjusted in real-time according to time-varying plant parameters. And the simulation result shows that the proposed control strategy is shown to be useful for the case of time-varying system and, robust to disturbances such as winds and initial sway motion.

• 산업경영시스템학회지
• /
• 제34권3호
• /
• pp.123-133
• /
• 2011

The advanced planning and scheduling(APS) is an well known enterprise information system that provides optimal production schedules and supports to complete production on time by solving the complex scheduling problems including capacity and due dates. In this paper, we focused on the improvement of the APS that is already established on a real company. The existing APS had several drawbacks, thus utilization and satisfaction were very low. We performed the focused group interviews and the process analysis and could find that the end users and developers have various objectives and the frequently used functions are different. We applied the analytical hierarchy process(AHP) to converge opinions of them on quantitative data. The results show that it is necessary to enhance visibility, to improve user interfaces and response speed, and to reconcile the real business process and the APS's process.

• 한국정보통신학회:학술대회논문집
• /
• 한국해양정보통신학회 2005년도 춘계종합학술대회
• /
• pp.573-577
• /
• 2005

현재 인터넷은 수요의 증가와 함께 사용되는 컨텐츠의 종류도 다양해지고 있다. 이런 다양한 컨텐츠의 트래픽들은 보다 좋은 질을 제공하기 위해 데이터양이 커지고 있으며, 성격이 상이한 부분이 많이 발생하고 있다. 하지만 현재의 인터넷은 BE(Best-Effort)서비스만을 제공하기 때문에, IP layer에서는 이러한 트래픽을 관리하고 QoS를 제공하기 위한 IntServ (Integrated Service)와 DiffServ (Differentiated Service)와 같은 기술들에 많은 연구가 진행되고 있다. 또한 무선인터넷의 증가와 함께 기존의 중요시 되지 못했던 MAC (Medium Access Control)의 중요성이 증가되고 있는데, 무선접속이라든지 자원할당에 있어 기존의 다른 계층에서 하던 상당수의 스케줄링이 MAC Layer에서 이루어지고 있다. 본 논문에서는 QoS를 제공하는 IEEE 802.11e EDCF 기반의 Mobile IPv6 망에서 트래픽의 특성에 맞는 QoS를 제공할 수 있는 알고리즘을 제안하고 시뮬레이션을 통해 검증한다.

• 한국차세대컴퓨팅학회논문지
• /
• 제15권4호
• /
• pp.18-29
• /
• 2019

동적 적응 스트리밍 서버는 일시에 많은 양의 트랜스코딩 연산을 처리하기 때문에 높은 프로세서 전력을 소모한다. 많은 연산량을 위하여 다중 프로세서 구조가 필요하고, 이에 대한 효과적인 트랜스코딩 태스크 분배가 필요하다. 본 논문에서는 2 티어 (프론트엔드 노드 (frontend node)와 백엔드 노드 (backend node)) 트랜스코딩 서버의 전력 상한을 보장하고 스트리밍 되는 비디오의 인기도 및 품질을 고려한 트랜스코딩 서버의 설계 및 구현 방법을 제안한다. 이를 위하여 1) 각 백엔드 노드에 트랜스코딩 태스크 분배, 2) 백엔드 노드에서의 태스크 스케줄링, 3) 프론트엔드와 백엔드 노드 통신 기법들을 구현하고, 테스트베드를 구축하였다. 실제 테스트베드에서의 예상 소모 전력과 실제 소모 전력을 비교하는 실험을 진행함으로써 본 시스템의 효용성을 확인했다. 또한 본 시스템이 각 노드의 부하를 감소시킴으로써 트랜스코딩에 사용되는 전력 및 시간 최적화가 가능함을 보였다.

• 한국정보과학회논문지:컴퓨팅의 실제 및 레터
• /
• 제15권2호
• /
• pp.89-97
• /
• 2009

유비쿼터스 환경에서 널리 사용되고 있는 내장형 시스템에서는 메모리 사용량의 감소와 신뢰성 등의 이유로 쓰레드 기반 프로그래밍 모델보다는 이벤트-구동형 프로그래밍 모델을 채용하는 경우가 많다. 그러나 내장형 시스템의 소프트웨어가 점점 더 복잡해지면서, 내장형 시스템을 위한 소프트웨어를 이벤트-구동형 프로그래밍 모델의 단일 이벤트 핸들러로 프로그램 하는 것은 매우 어려운 과제가 되고 있다. 본 논문에서는 내장형 시스템의 설계에 비선점형 스케줄링 기법을 도입하기 위하여, 다항시간의 스케줄 가능성 평가를 위한 충분조건과 이를 이용한 효율적인 스케줄 가능성 검사 방법을 제시하며, 또한 내장형 시스템에서의 서브-태스크의 활용이 비선점형 스케줄링을 채용했을 때의 단점인 낮은 처리기 이용률을 극복할 수 있는 방안임을 보인다.