| Summary: | 碩士 === 國立臺灣科技大學 === 資訊管理系 === 99 === With the rapid development of the Internet, the volume of network resources and resource accessing rate are not what we could imagine. It is even impossible to put all the data together in only one network file server, which not only reduces the efficiency but also raises the risk of unrecovered file damage. In addition, due to the rapid increasing of network users, once the request data cannot be serviced by nearby or not overloaded data server, users have to spend much time on waiting for server response. This situation causes many real time system cannot obtain essential data in specific time delay.
Recently, there are many researchers devoting to the research of using as least as possible replica servers to satisfy all the network users' Quality-of-Service (QoS) requirement. There are a lot of solutions by heuristic algorithms or approximation algorithms for similar research issues, but few of replica placement researches take the fault tolerance and replica server workload into consideration. In our research, we consider replica placement, fault tolerance and workload limits situation. We transform replica placement problem into network flow problem and find out the data flow deployment strategy with the minimum transmission cost. In addition, we use the architecture of Bipartite Graph integrates with network flow and find out lots of related problem solutions and improvements.
In this thesis, we will utilize graph algorithms to solve the problems or related applications after we transformed the replica placement problem into network flow model. Furthermore, we will propose an effective and near optimal replica placement heuristic algorithm. The experimental result also reveals that the difference between the optimal and our algorithm result is not obvious but the proposed algorithm is much faster than that of deriving optimal solutions.
|
|---|
