Font Size: a A A

Research On Cluster Management System

Posted on:2019-02-04Degree:MasterType:Thesis
Country:ChinaCandidate:X Z XuFull Text:PDF
GTID:2348330542498833Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
With the advancing development of the Internet technology,the amount of data shows a trend of explosive growth,so storing and analyzing these data is facing many challenges.It is difficult to analyze massive data through a single machine,distributed system has become an inevitable trend for mass data storage and analysis.A distributed system reduces the system's hardware costs by combining multiple inexpensive PC units into a single system that provides services to meet the needs of storing,computing and analyzing for massive data.With the inscreasement of the scale of data,the size of the cluster system is continuously expanding and the functions of the cluster system are becoming more complex,which increases the difficulty of cluster management.The reliability and consistency of the cluster system has become the key to the performance of cluster system.Therefore,how to provide a reliable and efficient management mechanism for cluster system is always one of the most important research topics.To reduce communication complexity and reduce the load of the system,this thesis deeply researches the traditional election algorithms and heartbeat detection models,proposes an election mechanism with low communication overhead,which reduces the communication complexity of the election process.The thesis also designs a heartbeat mechanism that combines the heartbeat information with the synchronization information to simplify the system structure and improve the network transmission efficiency.The main works are:(1)A low communication overhead election mechanism based on the node performance is discussed.The cluster system needs a master node to maintain cluster node information and assign external tasks.The election algorithm not only needs to elect the node with the highest performance to serve as the master node,but also needs to quickly select the optimal node to take over the work of the master node when the original master node fails.(2)An efficient bidirectional heartbeat detection mechanism is designed.The heartbeat mechanism is the basic module to ensure the failure detection of the cluster system.In order to improve the efficiency of network transmission,the thesis combines the synchronization of cluster information and heartbeat information in the heartbeat detection mechanism.(3)Key modules of the system are developed based on the election algorithm and the bidirectional heartbeat detection mechanism.The modules mainly include the election module,heartbeat module,machine node online processing module and the fault handling module.
Keywords/Search Tags:Cluster management, Election algorithm, Heartbeat detection
PDF Full Text Request
Related items