DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | Lee, Yoon-Joon | - |
dc.contributor.advisor | 이윤준 | - |
dc.contributor.author | Cho, Dae-Hyun | - |
dc.contributor.author | 조대현 | - |
dc.date.accessioned | 2011-12-13T05:26:43Z | - |
dc.date.available | 2011-12-13T05:26:43Z | - |
dc.date.issued | 2008 | - |
dc.identifier.uri | http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=295430&flag=dissertation | - |
dc.identifier.uri | http://hdl.handle.net/10203/33252 | - |
dc.description | 학위논문(박사) - 한국과학기술원 : 전산학전공, 2008.2, [ viii, 67 p. ] | - |
dc.description.abstract | The clustering of multiple video servers is a popular architecture for scalable VOD systems. In clustered video servers, user requests are distributed among the video servers by the dispatcher. Each video server contains independent disks and buffer memory. The buffer sharing mechanism is recognized helpful to the improvement of the throughputs of video servers. However, its effects have not been effectively utilized so far for the request distribution strategies of the clustered video servers. In this dissertation, we investigate the effects of buffer sharing on the throughput of the clustered video servers. And, based on the results, we propose two request distribution methods for the clustered video servers, where the buffer sharing effects are fully utilized for improving the cluster`s throughput. One method is designed for a static workload environment and another method for a dynamic workload environment. Both two proposed methods periodically build a distribution plan based on an estimated current workload and the analyzed buffer sharing effects, and then distribute incoming requests among video servers according to the plan. The plan specifies each server`s role: the number of requests for each video that the server will take. The method for static workload focuses on finding an ideal service state as a distribution plan whose throughput is the maximum among all possible service states for a given workload by maximizing buffer sharing. The experimental results for static workload show that our method can dramatically improve the cluster`s throughput by effectively utilizing the buffer sharing compared with the conventional request distribution methods. We also verify the optimality of the algorithm of finding an ideal service state. When the distribution plan is frequently updated under the dynamic workload, the cluster`s performance may be degraded due to temporarily non-optimal request assignments during the transition to a new plan. The... | eng |
dc.language | eng | - |
dc.publisher | 한국과학기술원 | - |
dc.subject | request distribution | - |
dc.subject | buffer sharing | - |
dc.subject | clustered video servers | - |
dc.subject | multimedia system | - |
dc.subject | 요청 분배 | - |
dc.subject | 버퍼 공유 | - |
dc.subject | 비디오 서버 클러스터 | - |
dc.subject | 멀티미디어 시스템 | - |
dc.subject | request distribution | - |
dc.subject | buffer sharing | - |
dc.subject | clustered video servers | - |
dc.subject | multimedia system | - |
dc.subject | 요청 분배 | - |
dc.subject | 버퍼 공유 | - |
dc.subject | 비디오 서버 클러스터 | - |
dc.subject | 멀티미디어 시스템 | - |
dc.title | Request distribution maximizing buffer sharing effects in clustered video servers | - |
dc.title.alternative | 비디오 서버 클러스터에서의 버퍼 공유 효과를 최대화하는 요청 분배 | - |
dc.type | Thesis(Ph.D) | - |
dc.identifier.CNRN | 295430/325007 | - |
dc.description.department | 한국과학기술원 : 전산학전공, | - |
dc.identifier.uid | 020005298 | - |
dc.contributor.localauthor | Lee, Yoon-Joon | - |
dc.contributor.localauthor | 이윤준 | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.