Request Shuffle : Novel Mechanisms to Improve Scalability of a Distributed Continuous Media Server

博士 === 國立成功大學 === 工程科學系 === 87 === Continuous media server is regarded as one of the most important components in new generation information system. Distributed continuous media server, which is one of the solutions to provide large scale service, involves a number of design challenges. Unfortunatel...

Full description

Bibliographic Details
Main Authors: Shiao-Li Tsao, 曹孝櫟
Other Authors: Yueh Min Huang
Format: Others
Language:en_US
Published: 1999
Online Access:http://ndltd.ncl.edu.tw/handle/85130765025420897435
Description
Summary:博士 === 國立成功大學 === 工程科學系 === 87 === Continuous media server is regarded as one of the most important components in new generation information system. Distributed continuous media server, which is one of the solutions to provide large scale service, involves a number of design challenges. Unfortunately, the previous studies only focused on a particular level of a distributed continuous media server, and some issues were ignored. In this paper, we propose a generic architecture of a distributed continuous media server system, which consists of a number of servers and clusters. We further investigate the load balancing problem and scalability in inter-cluster, intra-cluster and intra-server levels. We present a set of mechanisms, called request shuffle, to balance load between clusters and servers, and improve the service capacity and the scalability of a system. Request shuffle consists of two technologies, i.e. load shifting and request migration. Load shifting is applied to cluster level and it is used to improve the service capacity by dynamically adjusting the load of servers within a cluster. On the other hand, request migration is employed to server level and it is mainly to reduce the buffer requirement and initial delay while a server scales up. Moreover, we discuss the implementation techniques and implement the core of the proposed schemes. Both simulation and experimental results show that the proposed schemes can balance the load of servers and clusters, reduce the buffer requirement and initial delay, and achieve high scalability of a distributed continuous media server system.