• Proceedings of the Korea Information Processing Society Conference
• /
• 2003.05b
• /
• pp.1349-1352
• /
• 2003

웹 서버는 CPU time의 대부분인 $75{\sim}78%$를 시스템 코드에서 소비하며 사용자 코드에서는 생각보다 많은 시간을 소비하지 않는다. 이것은 웹 서버의 성능에 운영체제가 많은 영향을 끼치고 있음을 암시하는 것이다. 본 논문에서는 Linux Trace Toolkit를 이용하여, 하이브리드 멀티 프로세스 멀티 스래드 방식의 아파치 웹 서버가 구동 중인 동안 호출되는 시스템 호출의 동작 과정과 역할에 대해 알아보고 어떤 시스템 코드에서 오버해드가 큰지를 분석하였다.

• Proceedings of the Korean Society for Quality Management Conference
• /
• 2004.04a
• /
• pp.548-555
• /
• 2004

현재 변전소의 감시 및 제어 설비로 활용되고 있는 집중감시제어반은 24시간 상시 운전상태로 운영되기 때문에 컴퓨터의 CPU 및 메인보드, 하드디스크 등의 노후화로 인해 개개의 특성 및 기능이 저하되고 있다. 도입된 지 10년 가까이 지난 집중감시제어반에 대한 효율적 교체가 불가피하게 되었다. 이에 대한 최적 교체주기에 대한 합리적인 결정을 하기 위해서 신뢰분포함수와 평균사용일수(MTTF)를 구하였다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2000.04a
• /
• pp.89-91
• /
• 2000

멀티미디어 응용이 점점 널리 사용되면서 멀티미디어 응용에 적합한 태스크 모델의 연구가 진행되었으나, 기존의 태스크 모델은 멀티미디어 응용의 특성을 충분히 반영하지 못하였다. 본 논문에서는 멀티미디어 응용의 자원 사용량의 패턴에 기반한 확률적 멀티프레임 태스크 모델(PMF)을 제시하고, PMF를 멀티미디어 응용의 스케줄링에 적용하여 CPU 자원을 효율적으로 이용하면서 멀티미디어 응용에게 통계적 QoS를 제공할 수 있는 방안을 제시한다. 또한, 다양한 스케줄링 기법을 채용한 모의실험을 통해 제시한 태스크 모델이 자원을 최대한 활용 하면서 응용에게 QoS를 보장할 수 있음을 보인다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2002.11a
• /
• pp.501-504
• /
• 2002

실시간 제어 시스템은 그 특수성에 의해 개발자에게 상당한 제약이 있었다. 본 논문에서는 실시간 운영체제 중에서 Linux 를 기반으로 한 RTLinux 를 사용하였다. RTLinux를 StrongARM SA-1110 CPU가 탑재된 영상처리 보드에 올리는 과정을 기술하고, RTLinux의 module 이라는 작업 형태를 이용하여 로봇의 카메라에서 획득한 영상을 실시간으로 전송하도록 구현하였다. 또한, 사용자의 입력을 감시하는 부분을 실시간 모듈로 구현하여 제한된 시간 안에 원하는 결과를 산출하도록 구현하였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2007.11a
• /
• pp.644-647
• /
• 2007

많은 수의 CPU를 사용해 오랜 시간 계산하는 초대형 연산의 경우, 일부 노드나 통신회선의 장애로 연산 실패를 종종 겪는데, 이를 위해 응용 수준의 무정지 연산 시스템의 구현이 중요하다. 본 논문에서는 비동기 알고리즘을 사용한 이전 시스템의 약점을 보완하여, 동기 알고리즘에도 적용가능한 새로운 응용수준의 무정지 연산 시스템을 제안하고 선형시스템의 해법에 적용하였다.

• Journal of KIISE
• /
• v.41 no.10
• /
• pp.792-798
• /
• 2014

In spatial databases, R-tree is one of the most widely used indexing structures and many variants have been proposed for its performance improvement. Among these variants, Hilbert R-tree is a representative method using Hilbert curve to process large amounts of data without high cost split techniques to construct the R-tree. This Hilbert R-tree, however, is hardly applicable to large-scale applications in practice mainly due to high pre-processing costs and slow bulk-load time. To overcome the limitations of Hilbert R-tree, we propose a novel approach for parallelizing Hilbert mapping and thus accelerating bulk-loading of Hilbert R-tree on GPU memory. Hilbert R-tree based on GPU improves bulk-loading performance by applying the inversed-cell method and exploiting parallelism for packing the R-tree structure. Our experimental results show that the proposed scheme is up to 45 times faster compared to the traditional CPU-based bulk-loading schemes.

• Journal of KIISE
• /
• v.44 no.1
• /
• pp.13-20
• /
• 2017

A message processor is server software that receives various message formats from clients, creates the corresponding threads to process them, and lastly delivers the results to the destination. Considering that each function of an SEDA-based message processor has its own characteristics such as CPU-bound or IO-bound, this paper proposes a thread-mapping strategy called "FC-TM" (function-characteristic aware thread mapping) that schedules the threads to the cores based on the function characteristics in multi-core environments. This paper assumes that message-processor functions are static in the sense that they are pre-defined when the message processor is built; therefore, we profile each function in advance and map each thread to a core using the information in order to maximize the throughput. The benchmarking results show that the throughput increased by up to a maximum of 72 % compared with the previous studies when the ratio of the IO-bound functions to the CPU-bound functions exceeds a certain percentage.

• KIISE Transactions on Computing Practices
• /
• v.23 no.7
• /
• pp.417-423
• /
• 2017

Recently, many embedded devices that have the computing capability required for deep learning have become available; hence, many new applications using these devices are emerging. However, these embedded devices have an architecture different from that of PCs and high-performance servers. In this paper, we propose a method that improves the performance of deep-learning framework by considering the architecture of an embedded device that shares memory between the CPU and the GPU. The proposed method is implemented in Caffe, an open-source deep-learning framework, and is evaluated on an NVIDIA Jetson TK1 embedded device. In the experiment, we investigate the image recognition performance of several state-of-the-art deep-learning networks, including AlexNet, VGGNet, and GoogLeNet. Our results show that the proposed method can achieve significant performance gain. For instance, in AlexNet, we could reduce image recognition latency by about 33% and energy consumption by about 50%.

• The KIPS Transactions:PartA
• /
• v.14A no.7
• /
• pp.421-434
• /
• 2007

Among many other CG generated natural scenes, the representation of ocean surfaces is one of the most complicated and time-consuming problem because of its large extent and complex surface movement. We present a hybrid method to represent and animate unbound deep-water ocean surfaces by utilizing graphics processor as both simulation and rendering core. Our technique is mainly based on spectral approaches that generate a high-detailed height field using Fourier transform on a 2D regular grid. Additionally, we incorporate Gerstner model and generate low-detailed height field on a 2D projected grid in order to represent large waves and main structure of ocean surface. There is no interruption between CPU and GPU, and no need to transfer simulation results from the system memory to graphics hardware because the entire simulation and rending processes are done on graphics processor. As a result we can synthesize and render realistic water surfaces in real-time. Proposed techniques are readily adoptable to real-time applications such as computer games that have heavy work load on CPU but still demand plausible natural scenes.

• Journal of the Korea Computer Industry Society
• /
• v.5 no.3
• /
• pp.345-354
• /
• 2004

In these days, such services are popularized as E-commerce, E-government, multimedia services, and home networking applications. Most web traffics generated contemporarily basically use the Hyper Text Transfer Protocol(HTTP). Unfortunately, the HTTP is improper for these applications that comprise significant components of the web traffics. In this paper, we introduce a real-time contents compression architecture that maximizes the web service performance as well as reduces the response time. This architecture is built into the linux kernel-based web accelerating module. It guarantees not only the freshness of compressed contents but also the minimum time delay using an server-state adaptive algorithm, which can determine whether the server sends the compressed message considering the consumption of server resources when heavy requests reach the web server Also, We minimize the CPU overhead of the web server by exclusively implementing the compression kernel-thread. The testing results validates that this architecture saves the bandwidth of the web server and that elapsed time improvement is dramatic.