반사된 분산 저장 디스크 시스템에서 디스크 서비스 요구 메시지 전송이 블록될 확률 분석

具本根 충주대 산업과학기술연구소 1998 産業科學論文集 Vol.6 No.-

In our previous research for MDDS, we propose the MDDS, Mirrored Declustering Disk System, which is the disk array system for hypercube. And we also analyze the probabilistic model for disk service at node. But previous research do not provide the model for disk service time. In this paper we analyze the probability of blocking the message related to the disk service request. To analyze the probability of blocking, we analyze the blocking time at the link. The probability of blocking is used to analyze the disk service time.

이기종 클러스터를 위한 TSC 부하 분할 알고리즘

具本根 충주대 2003 産業科學論文集 Vol.11 No.-

Cluster is the cost-effective parallel processing system. In this paper, we propose the TSC load sharing algorithm for the heterogeneous cluster composed of nodes having the different configurations. To share the load, TSC algorithm uses the processor performance fo node.Therefore, TSC algorithm effectively distributes the load to the nodes in the heterogeneous cluster. To show the effectiveness of TSC algorithm, we write the parallel programs for the matrix multiplication and the Mandelbrot set. And we compare the TSC with other load sharing algorithms-Send, GSS. The results of our experiments shows that TSC is more effective load sharing algorithm than Send, GSS. The space-time diagrams of parallel programs show that TSC distributes the load to nodes fairly. Therefore, TSC load sharing algorithm can enhance the performance of heterogeneous cluster by effectively sharing the load.

멀티미디어 서버를 위한 고성능 디스크 시스템

구본근 忠州大學校 1999 한국교통대학교 논문집 Vol.34 No.2

In this research we propose the RAID level-5g disk system for multimedia server. RAID level-5g consists of the disk groups. Each disk grop consists of the disks, and stores one media, such as moving pictures, sound, text, and etc. The disk access at each disk grop is independent of the disk access at other disk groups. Therefore disk access for multimedia data, consisting of multiple media data, are concurrently served in RAID level-5g. Because RAID level-5g can concurrently serve the disk access for many multimedia data request of the multimedia clients, it is possible to play the multimedia data with real-time manner in the clients.

클러스터 상에서의 적응적 부하 분할 정책

구본근 忠州大學校 2002 한국교통대학교 논문집 Vol.37 No.3

Cluster is the cost-effective parallel processing environment, and consists of the off-the-shelf computers connected by the computer networks. The characteristics of cluster are the node heterogeneity, the variety of node load, and the variety of network load. Because these characteristics influence the performance of parallel program executions, the load sharing for cluster is important, and by using the proper load sharing strategy, we can reduce the execution time of parallel programs. In this paper, we propose modified GSS algorithm, aGSS. In the proposed load sharing algorithms aGSS, the size of tasks are decided using the BogoMIPS of node. From the result of out experiments, we conclude that the proposed aGSS algorithm is effective in the heterogeneous cluster.

사용자 시선을 이용한 상황 정보

구본근 충주대학교 2006 한국교통대학교 논문집 Vol.41 No.-

In this paper, we propose the gaze direction context information. In many context-aware system, the services based on the user's locational information are provided. But we think that user firstly gazes at the object in which he is interested, and then he goes to that object. Therefore we propose and define the concept of the gaze proximity context. This context information is used to decide the service which the context-aware system provides user with. The context-aware system determines the gaze proximity context using the user's gaze direction and location information.

이기종 클러스터를 위한 TSCCP 부하 분할 알고리즘

具本根 충주대 2004 産業科學論文集 Vol.12 No.-

Cluster is the cost-effective parallel processing system. In this paper, we propose the TSCCP load sharing algorithm for the heterogeneous cluster composed of nodes having different configurations. To share the load, TSCCP algorithm uses the processor performance of node. Therefore, TSCCP algorithm effectively distributes the load to the nodes in the heterogeneous cluster. To show the effectiveness of TSCCP algorithm, we write two parallel programs for the matrix multiplication and the Mandelbrot set. And we compare the TSCCP with other load sharing algorithms―Send, GSS. The results of our experiments show that TSCCP is more effective load sharing algorithm than Send, GSS. And, we show that TSCCP is especially more effective for the computing intensive applications.

NOW 환경에서 preSend 알고리즘의 Java 구현

具本根 충주대 2001 한국교통대학교 논문집 Vol.36 No.2

In this paper, we propose the data server - calculation server model.In this model, the data server provides the data to the calculation server, and collects the results received from the calculation server.The calculation server performs the calculation using the data received from the data server, and returns the results of the calculations.We implement the proposed model using Java.We use the Java RMI(Remote Method Invocation) for exchanging the data and result values between the data server and the calculation server.The implementation in the paper adds the integers between 1 and 99,999 in NOW.The NOW in this paper consists two computers.We use MS-Windows 98, Linux and, JDK 1.3.1.We find that the proposed model is efficiently implemented using Java and Java RMI.

NOW환경에서 효율적인 부하 분할

具本根 충주대 2000 한국교통대학교 논문집 Vol.35 No.2

At NOW(Networks Of Workstations), the load sharing is very important role for improving the performance.The known load sharing strategy is fixed-granularity, variable-granularity, adaptive-granularity.The send algorithm, which implements the flexed-granularity strategy, gets higher performance than other algorithms on NOW.But, in Send algorithm, the computing time and the communication time are not overlapped.Therefore, long latency time at the network has influence on the execution lime of the parallel program.In this paper, we propose the preSend algorithm.In the preSend algorithm, the computing time and the communication time are overlapped.Therefore we reduce the influence of the long latency time at the network and enhance the performance of the NOW.

이기종 클러스터에서 부하 분할 모델

구본근 忠州大學校 2004 한국교통대학교 논문집 Vol.39 No.1

Cluster is the cost-effective parallel processing system. In this paper, we propose the Load-Sharing Model for the heterogeneous cluster composed of nodes having different configurations. To share the load, our load sharing model uses the processing time of slaves. The processing time of slave consists of computing time of task and communication time between master and slave. In our model, we use the gap between the processing time point of slave and LB point to load sharing. For the Load-Sharing Model, we define the processing time graph, processing time point, degree of imbalance. And we also describe the partition method of task and load sharing algorithm.

부분 중복 저장 기법을 이용한 디스크 시스템

具本根 충주대 산업과학기술연구소 1999 産業科學論文集 Vol.7 No.1

In this paper, we propose the new disk system, ML3, for the multiprocessor systems. ML3 is recursively configured with 3-dimensional hypercube architecture. ML31, called the node group, consists of eight P(Processor)-node and one D(Disk)-node. D-node serves all disk access for P-nodes. And ML32 consists of eight ML31 and one D-node. Disks in D-node of ML31 store data using declustering technique. Disks in D-node of ML32 duplicate the contents of disks in B-nodes of ML31. By using this manner, we can expand ML3n disk system. And we describe the process of disk service. In ML3, the message traffic is clustered at D-node. But we solve this problem by using the threshold of link availability.