• Proceedings of the Korea Information Processing Society Conference
• /
• 2011.04a
• /
• pp.11-13
• /
• 2011

최근 스마트 폰, 타블렛 PC 등 고성능 모바일 기기에 대한 시장수요가 증가함에 따라 임베디드 프로세서에 대한 성능 최적화가 활발히 이루어지고 있다. 고성능 멀티미디어 시스템을 대상으로 설계된 Empress 프로세서는 다수의 기능 유닛을 포함하고 있어 명령어 스케줄링을 통해 성능 향상을 기대할 수 있으나 기존의 컴파일러는 이를 지원하지 않고 있다. 본 연구에서는 GNU C 컴파일러를 이용하여 Empress 프로세서를 위한 DFA 기반의 명령어 스케줄링 최적화를 구현하였다. 그 결과 EEMBC 벤치마크를 이용한 성능 분석에서 실행시간 기준 평균 8%의 향상이 있음을 확인하였다.

• Journal of the Korea Society for Simulation
• /
• v.32 no.3
• /
• pp.67-74
• /
• 2023

A semiconductor fabrication facility(FAB) is one of the most capital-intensive and large-scale manufacturing systems which operate under complex and uncertain constraints through hundreds of fabrication steps. To improve fab performance with intuitive scheduling, practitioners have used weighted-sum scheduling. Since the determination of weights in the scheduling significantly affects fab performance, they often rely on simulation-based decision making for obtaining optimal weights. However, a large-scale and high-fidelity simulation generally is time-intensive to evaluate with an exhaustive search. In this study, we investigated three sampling methods (i.e., Optimal latin hypercube sampling(OLHS), Genetic algorithm(GA), and Decision tree based sequential search(DSS)) for the optimization. Our simulation experiments demonstrate that: (1) three methods outperform greedy heuristics in performance metrics; (2) GA and DSS can be promising tools to accelerate the decision-making process.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.16 no.3
• /
• pp.137-144
• /
• 2016

In this paper, we have proposed a circuit complexity optimization ILP algorithm of high-level synthesis system for new multiprocessor design. We have analyzed to the operator characteristics and structure of datapath in the most important high-level synthesis. We also introduced the concept of virtual operator for the scheduling of multi-cycle operations. Thus, we demonstrated the complexity to implement a multi-cycle operation of the operator, regardless of the type of operation that can be applied for commonly use in the ILP algorithm. We have achieved is that standard benchmark model for the scheduling of the 5th digital wave filter, it was exactly the same due to the existing datapath scheduling results.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.24 no.3
• /
• pp.63-67
• /
• 2024

Recently, artificial intelligence (AI) workloads encompassing various industries such as smart logistics, FinTech, and entertainment are being executed on the cloud. In this paper, we address the scheduling issues of various AI workloads on a multi-tenant cloud system composed of heterogeneous GPU clusters. Traditional scheduling decreases GPU utilization in such environments, degrading system performance significantly. To resolve these issues, we present a new scheduling approach utilizing genetic algorithm-based optimization techniques, implemented within a process-based event simulation framework. Trace driven simulations with diverse AI workload traces collected from Alibaba's MLaaS cluster demonstrate that the proposed scheduling improves GPU utilization compared to conventional scheduling significantly.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2012.04a
• /
• pp.10-13
• /
• 2012

임베디드 시스템은 여러 분야에서 사용되고 있으며, 그 범위는 더욱더 다양하게 늘어나고 있다. 이러한 다양성은 임베디드 시스템이 사용되는 목적에 따라 새로운 아키텍처를 요구하게 되면서, 아키텍처 구조, 동작에 대한 변경 또는 새로운 설계에 대해 개발 시간과 비용을 줄이기 위한 재목적 컴파일러의 개발 필요성과 중요성이 강조되고 있다. 더욱이 전력이 제한적인 모바일 기기에서 동작하는 어플리케이션의 최적화와 이러한 최적화를 위한 컴파일러 개발은 매우 중요한 이슈가 되고 있으며, 특히 어플리케이션 성능에 직접적인 영향을 주는 컴파일러 후단부는 다양한 방법론들이 적용되어 있고 많은 연구가 수행되고 있다. 이 논문에서는 EXPRESSION의 재목적 컴파일러인 EXPRESS의 후단부에서 코드 최적화를 위해 적용된 기법을 분석하고, 기존 코드 스케줄링과 더불어 성능 개선을 위해서 기본 블록 스케줄링을 추가한 모델을 설계하고 성능평가 방법을 제시한다.

• KIPS Transactions on Computer and Communication Systems
• /
• v.3 no.5
• /
• pp.151-154
• /
• 2014

An energy of applications had consumed through the complexed inter-action with operating systems, run-time environments, compiler, and applications on various mobile devices. In these days, challenged researches are studying to reduce of energy consumptions that uses energy-oriented high-level and low-level compiler techniques on mobile devices. In this paper, we intented to reduce an energy consumption of Java mobile applications that applied a list instruction scheduling for energy dissipation from dalvik bytecode which extracted Android dex files. Through this works, we can construct the optimized power and energy environment on mobile devices with the limited power supply.

• Journal of Korea Society of Industrial Information Systems
• /
• v.8 no.3
• /
• pp.19-31
• /
• 2003

For the implementation of large scale GRID systems, the performance scalability in resource scheduling is clearly to be addressed. In this research, we analyzed existing scheduling frameworks from the viewpoint of the performance and propose a novel resource scheduling framework called resource parameter based scheduling. Proposed scheduling framework consists of three components. The first is the resource parameter graph that expresses resource information via inter-resource relation and the composition base on the hierarchical structure. The second component is the resource parameter tree to be used for the implementation of the memory-based index of resource information. The third component is the resource information repository which mostly consists of static data to be used for the general resource information services. This paper presents the details of the framework.

• Proceedings of the Korean Information Science Society Conference
• /
• 2012.06a
• /
• pp.282-284
• /
• 2012

임베디드 분야에서 전력 에너지 소비 문제는 시스템을 설계하는데 있어서 매우 중요한 이슈가 되고 있다. 특히 휴대성이 강조되는 모바일 장치의 제한된 전력을 효율적으로 이용하기 위해서 하드웨어적인 관리 못지않게 소프트웨어적인 관리 기술의 필요성이 강조되고 있으며 전력 소비 관리를 위한 최적화된 컴파일러 기법이 연구되고 있다. 이 논문에서는 모바일 장치에서 구동되는 어플리케이션의 전력 에너지 소비를 줄이기 위한 전역 코드 스케줄링 기법을 제시한다. 이를 위해, 재목적 소프트웨어 개발 도구인 EXPRESSION의 컴파일러인 EXPRESS의 코드 최적화 기법을 이용하여 전력 에너지 효율적인 전역 코드 스케줄링 모델을 설계하고 성능평가 방법을 제시한다.

• Journal of Internet Computing and Services
• /
• v.12 no.6
• /
• pp.1-17
• /
• 2011

The performance of DHCS depends on the algorithm which schedules input DAG. However, as the task scheduling problem in DHCS is an NP-complete problem, heuristic approach has to be made. Task scheduling algorithm consists of task prioritizing phase and processor allocation phase, and most of studies are considering both phases together. In this paper, we focus on task prioritizing phase and propose a WPD algorithm which is optimized for task duplication based processor allocation method. For an evaluation of the proposed WPD algorithm, we combined WPD algorithm with processor allocation phase of HMPID, HCPFD, HCT algorithms, which are using task duplication based processor allocation method. The results show that WPD algorithm makes a better use of task duplication than conventional task prioritizing methods and provides 9.58% better performance than HCPFD algorithm, 1.31% than HCT algorithm.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2001.10a
• /
• pp.113-116
• /
• 2001

In this paper, a new multicasting scheduling is proposed for WDM single-hop broadcast network. The main idea of traditional multicast scheduling algorithms is partitioning the multicast group into several subgroups to reduce the delay time after receiving previous message. Those partitioning algorithms are grouping method of receivers already tuned to transmitters wavelength. However, multicast scheduling shows better performance by considering the previous status of receivers. Thus, in this paper, a new subgrouping method is proposed to minimize delay through collecting the current status of receivers. This method is operated by using the tunable transceiver in WDM single hop broadcasting network and uses fixed control channel for collecting the status of receivers. In addition, we analyzed the performance of proposed method through the computer simulation and then compared with existing multicast scheduling algorithm.