• Proceedings of the Korea Information Processing Society Conference
• /
• 2004.05a
• /
• pp.127-130
• /
• 2004

본 논문에서는 경성 실시간 환경에서 MPEG 비디오 스트림을 디코딩하는 멀티미디어 태스크와 경성 실시간 태스크를 효율적으로 스케쥴링하는 기법이 제시된다. 연성 실시간 특성을 갖는 멀티미디어 태스크는 평균 실행 시간을 기반으로 처리되는 반면 경성 실시간 태스크는 최악의 경우에 대한 실행 시간을 기반으로 실행이 보장된다. 서버를 기반으로 하는 본 논문의 스케쥴링 기법은 CPU 대역폭을 두 태스크들에 대하여 분할하며 멀티미디어 태스크들에 대하여는 할당된 대역폭이 다시 동적으로 조정한다. 제시된 기법의 목적은 시스템에 존재하는 경성 실시간 태스크들의 시간적 제약성을 모두 보장하면서 멀티미디어 태스크들의 종료시한이후 실행이 지연되는 시간을 최소화하는 것이다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2013.11a
• /
• pp.135-138
• /
• 2013

GPGPU(General Purpose Graphics Processing Unit) 병렬처리 시스템인 CUDA(Compute Unified Device Architecture)는 컴퓨터에서의 고속 연산 처리를 위해 많이 사용되어왔다. CUDA에서 연산 처리를 하기 위해서는 CUDA의 특성을 이해해야 한다. CUDA는 CPU(Central Processing Unit)가 처리하는 Host 영역과 GPU(Graphics Processing Unit)가 처리하는 영역인 Device 영역이 존재하며, 이 두 영역간의 데이터 복사를 통해 연산 처리를 진행한다. 이런 구조적인 특성상 메인 메모리에서 GPU 메모리로 입력 데이터를 전달해야 GPU를 이용해 연산을 처리할 수 있는 구조를 가지고 있다. 하지만 이러한 처리 구조로 인해 연산 시간과 별도로 메인 메모리와 GPU 메모리간의 데이터 복사시간이 존재하며, 추가적으로 발생하는 메모리 복사 시간으로 인해 오버헤드가 발생하게 된다. 본 논문에서는 실험을 통해 메모리 복사 시간, 연산의 반복 횟수 그리고 연산의 복잡성이 전체 성능에 어떤 영향을 미치는지 논하고자 한다.

• Journal of KIISE:Computing Practices and Letters
• /
• v.15 no.7
• /
• pp.510-514
• /
• 2009

If the existing real-time systems are integrated based on SOA, real-time SOA should be developed. Generally, in real-time SOA a service can be divided into several small services and their estimated execution time is given by provider systems. However, an estimation, which analyzes time elements related to transmit and receive messages among requesters and providers, is needed. In order to enhance QoS of Web service, this paper proposes enhanced worst-case execution time estimation by considering WS-transaction and common use of multi-processors system.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2008.05a
• /
• pp.734-737
• /
• 2008

This paper presents a new automatic TFT-LCD image quality optimization system. We also have developed new algorithms using 6-point programmable matching technique with reference gamma curve, and automatic power setting sequence. It optimizes automatically gamma adjustment and power setting registers in mobile TFT-LCD driver IC to reduce gamma correction error, adjusting time, and flicker. Developed algorithms and programs 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 and flicker, and control board for interface between PC and TFT-LCD module. The control board is designed with DSP, and it supports various interfaces such as RGB and CPU. Developed automatic image quality optimization system showed significantly reduced gamma adjusting time, reduced flicker, and much less average gamma error than competing system. We believe that the proposed system is very useful to provide high image quality TFT-LCD and to reduce developing process time using optimized gamma-curve setting and automatic power setting.

• 전자공학회논문지 IE
• /
• v.45 no.1
• /
• pp.15-23
• /
• 2008

In this paper, an algorithm, that provisions absolute differentiation of packet delays is proposed, simulated, and implemented with VHDL on XPC 860 CPU based test board with an objective for enhancing quality of service (QoS) in future packet networks. It features a scheme that compensates the deviation for prediction on the traffic to be arrived continuously. It predicts the traffic to be arrived at the beginning of a time slot and measures the actual arrived traffic at the end of the time slot and derives the difference between them. The deviation is utilized to the delay control operation for the next time slot to offset it. As it compensates the prediction error continuously, it shows superior adaptability to the bursty traffic as well as the exponential traffic. It is demonstrated through both simulation and the real traffic test on the board that the algorithm meets the quantitative delay bounds and shows superiority to the traffic fluctuation in comparison with the conventional non-adaptive mechanism.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.14 no.6
• /
• pp.75-84
• /
• 2014

In order to solve the problems such as overload, single point of failure, non-optimized data path, and network scalability in conventional central mobility management protocols, distributed mobility management schemes have been continually studied in and around the IETF. In this paper, a network-based partially-distributed mobility management mechanism, pDMMv6, is suggested and the performance comparison with traditional protocols such as PMIPv6 and MIPv6 is made through simulation under the various user traffic environment. The simulation results include UDP packet delivery ratio, end-to-end packet delay, binding delay for registration signaling, CPU utilization in each node, and response delays in several server-client TCP applications such as web browsing, e-mail, telnet remote login, FTP file up/down-load, and database access.

• Convergence Security Journal
• /
• v.7 no.2
• /
• pp.17-25
• /
• 2007

In this paper, an algorithm that provisions proportional differentiation of packet delays is proposed with an objective for enhancing quality of service (QoS) in future packet networks. It features an adaptive scheme that adjusts the target delay every time slot to compensate the deviation from the target delay which is caused by the prediction error on the traffic to be arrived in the next time slot. It predicts the traffic to be arrived at the beginning of a time slot and measures the actual arrived traffic at the end of the time slot. The difference between them is utilized to the delay control operation for the next time slot to offset it. As it compensates the prediction error continuously, it shows superior adaptability to the bursty traffic as well as the exponential rate traffic. It is demonstrated through simulations that the algorithm meets the quantitative delay bounds and shows superiority to the traffic fluctuation in comparison with the conventional non-adaptive mechanism. The algorithm is implemented with VHDL on a Xilinx Spartan XC3S1500 FPGA and the performance is verified under the test board based on the XPC860P CPU.

• Journal of Digital Contents Society
• /
• v.15 no.5
• /
• pp.631-642
• /
• 2014

The main memory DBMS is operated which the contents of the table that resides on a disk at the same time as the drive is in the memory. However, because the main memory DBMS stores the data and transaction log file using the disk file system, there are a limit to the speed at which the CPU accesses the memory. In this paper, I evaluated the performance through analysis of the application side difference the technology that has been implemented in Altibase system of main memory DBMS and Sybase of disk-based DBMS. When the application performance of main memory DBMS is in comparison with the disk-based DBMS, the performance of main memory DBMS was outperformed 1.24~3.36 times in the single soccer game, and was outperformed 1.29~7.9 times in the soccer game / special soccer. The result of sale transaction response time showed a fast response time of 1.78 ~ 6.09 times.

• Journal of the Korea Society of Computer and Information
• /
• v.15 no.9
• /
• pp.99-107
• /
• 2010

This paper showed that the integrated system to fortify security was much more efficient than the respective system through the analysis of problems from Firewall and IPS system in the existing security systems. The results of problem analysis revealed that there were the delay of processing time and lack of efficiency in the existing security systems. Accordingly, their performance was evaluated by using the separated Firewall, IPS system, and the integrated system. The result of evaluation shows that the integrated security system this paper suggested is five times faster than the existing one in terms of processing speed of response. This paper demonstrated the excellence of the proposed security system is also more than fivefold in session handling per second and six times process speeding in the CPU processing performance. In addition, several security policies are applied, and it provided a fact that it gave an excellent performance when it comes to protecting from harmful traffic attacks. In conclusion, this paper emphasized that fortifying the integrated security system was more efficient than fortifying the existing one considering in various respects such as cost, management, time, space and so on.

• KIPS Transactions on Computer and Communication Systems
• /
• v.11 no.7
• /
• pp.205-216
• /
• 2022

One of the many tools used for distributed deep learning training is Kubeflow, which runs on Kubernetes, a container orchestration tool. TensorFlow jobs can be managed using the existing operator provided by Kubeflow. However, when considering the distributed deep learning training jobs based on the parameter server architecture, the scheduling policy used by the existing operator does not consider the task affinity of the distributed training job and does not provide the ability to dynamically allocate or release resources. This can lead to long job completion time and low resource utilization rate. Therefore, in this paper we proposes a new operator that efficiently schedules distributed deep learning training jobs to minimize the job completion time and increase resource utilization rate. We implemented the new operator by modifying the existing operator and conducted experiments to evaluate its performance. The experiment results showed that our scheduling policy improved the average job completion time reduction rate of up to 84% and average CPU utilization increase rate of up to 92%.