【摘要】 为解决Ceph默认参数无法充分发挥系统读写性能，手动调整参数效率低下、浪费大量系统资源的问题，提出一种基于长短期记忆网络和精英保留遗传算法的Ceph参数自动调优方法。采集真实环境下不同参数组合所对应的系统读写性能，构成实验所需的数据集，在此基础上通过LSTM构建Ceph性能预测模型，使用精英保留遗传算法寻找最优的参数组合，减少在真实环境中测试所消耗的时间和系统资源。通过实验，验证了该方法在准确率、收敛速度和性能提升等方面优于现有方法，经过调优后的系统读写性能是默认参数的1.7倍。更多还原

【Abstract】 The default parameters of Ceph can not give full play to the read and write performance of the system, the manual adjustment of parameters is inefficient and a lot of system resources are wasted. Therefore, an automatic tuning method of Ceph parameters based on long short-term memory network and elitism genetic algorithm was proposed. The system read-write performance corresponding to different parameter combinations in the real environment was collected to form the data set required for the experiment. On this basis, the CEPH performance prediction model was constructed through LSTM, and the elite rese-rvation genetic algorithm was used to find the optimal parameter combination, so as to reduce the time and system resources consumed in the real environment. Through experiments, it is verified that this method is superior to the existing methods in accuracy, convergence speed and performance improvement, and the read-write performance of the optimized system is 1.7 times better than that of the default parameters.更多还原