Research And Implementation Of Key Technologies On High-Availability Cluster System

Nowadays, the computer system enters an era of cloud computing and big data, with inevitablyusing large-scale server cluster responding to mass requests and processing mass data.High-performance cluster system cannot meet the demand of these large-scale applications. One offocused problems in current cluster system researches is how to improve the availability of clustersystem, for avoiding the serious consequences caused by system failure and designing an extensiblehigh-availability system.This thesis focused on the high-availability cluster system, designed a framework for thehigh-availability cluster system and further studies its two critical problems-fault tolerancemechanism and task allocation&migration mechanism. Two algorithms are proposed herein:double-layered heartbeat detection algorithm based on ARMA prediction and load balancingalgorithm with key resource relevance. The heartbeat detection algorithm predicts the arrival timeinterval of heartbeat packets using ARMA and dynamically sets the time-out of heartbeat packets, andthus reduces the wrong judgment of node states caused by fixing time-out threshold of heartbeatpackets. In contrast with the current load balancing algorithm without focuses on the task relevance toresources, the new load balancing algorithm herein fully considers the relevance of tasks to each noderesources, which can allocate the task to the high-relevance node while assuring the system loadbalance.In the last part of this thesis, we used the heartbeat detection algorithm for the fault tolerancemechanism and the load balancing algorithm for task allocation mechanism. We implemented thecore-part of the high-availability cluster system and test the system’s performance through series ofexperiments. The experiments show that: the system has high fault detection accuracy, excellent loadbalancing performance and short task waiting time.