• The Journal of the Korea institute of electronic communication sciences
• /
• v.14 no.6
• /
• pp.1207-1214
• /
• 2019

Technologies of the 4th industrial revolution have been unknowingly seeping into our lives. Many IoT based home security systems are using the convolutional neural network(CNN) as good biometrics to recognize a face and protect home and family from intruders since CNN has demonstrated its excellent ability in image recognition. In this paper, three layouts of CNN for 2D and 2.5D image of small dataset with various input image size and filter size are explored. The simulation results show that the layout of CNN with 50*50 input size of 2.5D image, 2 convolution and max pooling layer, and 3*3 filter size for small dataset of 2.5D image is optimal for a home security system with recognition accuracy of 0.966. In addition, the longest CPU time consumption for one input image is 0.057S. The proposed layout of CNN for a face recognition is suitable to control the actuators in the home security system because a home security system requires good face recognition and short recognition time.

• Journal of KIISE:Computer Systems and Theory
• /
• v.33 no.3
• /
• pp.193-204
• /
• 2006

In most existing distributed Web systems, incoming requests are distributed to servers via Domain Name System (DNS). Although such systems are simple to implement, the address caching mechanism easily results in load unbalancing among servers. Moreover, modification of the DNS is necessary to load considering the server's state. In this paper, we propose a new dynamic load balancing method using dynamic DNS update and round-robin mechanism. The proposed method performs effective load balancing without modification of the DNS. In this method, a server can dynamically be added to or removed from the DNS list according to the server's load. By removing the overloaded server from the DNS list, the response time becomes faster. For dynamic scheduling, we propose a scheduling algorithm that considers the CPU, memory, and network usage. We can select a scheduling policy based on resources usage. The proposed system can easily be managed by a GUI-based management tool. Experiments show that modules implemented in this paper have low impact on the proposed system. Furthermore, experiments show that both the response time and the file transfer rate of the proposed system are faster than those of a pure Round-Robin DNS.

• Journal of KIISE
• /
• v.44 no.6
• /
• pp.566-572
• /
• 2017

As NAND flash storage is being widely used, its simulation methodologies have been studied in various aspects such as performance, reliability, and endurance. As a result, there have been advances in NAND flash storage simulation for both functional modeling and timing modeling. However, in addition to these advances, there is a need to drastically reduce the long simulation time that is required to evaluate the aging effect on flash storage. This paper proposes a so-called multi-core scalable real-time flash storage simulation method, which can control the simulation speed according to the user's preference. According to this method, it is possible to speed up the simulation in proportion to the number of CPU cores arbitrarily given while guaranteeing the correctness of the simulation result. Using our simulator implemented in the form of the Linux kernel module, we demonstrate the multi-core scalability and correctness of the proposed method.

• Journal of Korea Game Society
• /
• v.13 no.5
• /
• pp.55-62
• /
• 2013

Since large terrain data can not be loaded on the GPU or CPU memory at once, out-of-core methods which read necessary part from the secondary storage such as a hard disk are commonly used. However, long delay may occur due to limited bandwidth while loading the data from the hard disk to memory. We propose efficient rendering method of large terrain data, which compresses the data with wavelet technique and save its coefficients in RGBA channel of an image us, then decompresses that in rendering stage. Entire process is performed in GPU using Direct Compute. By reducing the amount of data transfer, performing wavelet computations in parallel and doing decompression quickly on the GPU, our method can reduce rendering time effectively.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2008.10a
• /
• pp.592-595
• /
• 2008

This paper presents development of automatic image quality register optimization system using mobile TFT-LCD (Thin Film Transistor-Liquid Crystal Display) driver IC and embedded software. It optimizes automatically gamma adjustment and voltage setting registers in mobile TFT-LCD driver IC to improve gamma correction error, adjusting time, flicker noise and contrast ratio. Developed algorithms and embedded software are generally applicable for most of the TFT-LCD modules. The proposed optimization system contains module-under-test (MUT, TFT-LCD module), control program, multimedia display tester for measuring luminance, flicker noise and contrast ratio, and control board for interface between PC and TFT-LCD module. The control board is designed with DSP and FPGA, and it supports various interfaces such as RGB and CPU.

• KIPS Transactions on Computer and Communication Systems
• /
• v.3 no.7
• /
• pp.209-218
• /
• 2014

An AP that is used by smart device is going to be complicated because smart devices support diverse functions. As a result, multiple low-level IPs including a dedicated CPU are integrated in a high-level subsystem for supporting complicated function such as multimedia codec and camera. A subsystem has software that executes separately from main system, and the software needs to be initialized for every execution of the subsystem. This causes increase of the subsystem startup time so it should be improved because startup time of subsystem affects launching time of application. Methods in applied to computer system for fast startup also could be applied to fast startup of subsystem because subsystem is similar with computer system. In this paper, we apply snapshot method that is used in computer system to subsystem and analyzes the pros and cons. And snapshot method could not be applied to register of IP without modification because register of IP offers restricted read and write. So this paper suggests technique that applying snapshot to each characteristic of register.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2017.05a
• /
• pp.335-338
• /
• 2017

Currently Artificial Inteligence and Deep Learning are social issues, and These technologies are applied to various fields. A good method among the various algorithms in Artificial Inteligence is Convolutional Neural Network. Convolutional Neural Network is a form that adds convolution layers that extracts features by convolution operation on a general neural network method. If you use Convolutional Neural Network as small amount of data, or if the structure of layers is not complicated, you don't have to pay attention to speed. But the learning time is long as the size of the learning data is large and the structure of layers is complicated. So, GPU-based parallel processing is a lot. In this paper, we developed Convolutional Neural Network using CUDA and Learning speed is faster and more efficient than the method using the CPU.

• The KIPS Transactions:PartA
• /
• v.12A no.2 s.92
• /
• pp.151-160
• /
• 2005

The exponential growth of Web users requires the web serves with high expandability and reliability. It leads to the excessive transmission traffic and system overload problems. To solve these problems, cluster systems are widely studied. In conventional cluster systems, when the request size is large owing to such types as multimedia and CGI, the particular server load and response time tend to increase even if the overall loads are distributed evenly. In this paper, a cluster system is proposed where each Web server in the system has different contents and loads are distributed efficiently using the Web server resource information such as CPU, memory and disk utilization. Web servers having different contents are mutually connected and managed with a network file system to maintain information consistency required to support resource information updates, deletions, and additions. Load unbalance among contents group owing to distribution of contents can be alleviated by reassignment of Web servers. Using a simulation method, we showed that our method shows up to $50\%$ about average throughput and processing time improvement comparing to systems using each LC method and RR method.

• Journal of KIISE:Software and Applications
• /
• v.31 no.2
• /
• pp.164-172
• /
• 2004

For a local search algorithm to find a bettor quality solution it is required to generate and evaluate a sufficiently large number of candidate solutions as neighbors at each iteration, demanding quite an amount of CPU time. This paper presents a method of selectively generating only good-looking candidate neighbors, so that the number of neighbors can be kept low to improve the efficiency of search. In our method, a newly generated candidate solution is probabilistically selected to become a neighbor based on the quality estimation determined heuristically by a very simple evaluation of the generated candidate. Experimental results on the problem of load balancing for production scheduling have shown that our candidate selection method outperforms other random or greedy selection methods in terms of solution quality given the same amount of CPU time.

• Journal of Korea Water Resources Association
• /
• v.52 no.11
• /
• pp.887-894
• /
• 2019

We proposed a GPU (Grapic Processing Unit) accelerated kinematic wave model for rainfall runoff simulation and tested the accuracy and speed up performance of the proposed model. The governing equations are the kinematic wave equation for surface flow and the Green-Ampt model for infiltration. The kinematic wave equations were discretized using a finite volume method and CUDA fortran was used to implement the rainfall runoff model. Several numerical tests were conducted. The computed results of the GPU accelerated kinematic wave model were compared with several measured and other numerical results and reasonable agreements were observed from the comparisons. The speed up performance of the GPU accelerated model increased as the number of grids increased, achieving a maximum speed up of approximately 450 times compared to a CPU (Central Processing Unit) version, at least for the tested computing resources.