摘要:云计算已成为大数据分析作业的主流运行支撑环境，选择合适的云资源优化其性能面临巨大挑战.当前研究主要考虑大数据分析框架（如Hadoop，Spark等）的多样性，采用机器学习方法进行资源供给，但样本少容易陷入局部最优解.提出了大数据环境下基于负载分类的启发式云资源供给方法RP-CH，基于云资源共享特点，获取其他大数据分析作业的运行时监测和云资源配置信息，建立负载分类与优化云资源配置的启发式规则，并将该规则作用到贝叶斯优化算法的收益函数.基于HiBench，SparkBench测试基准的结果显示：RP-CH相对于已有方法CherryPick、大数据分析作业的性能平均提升了58%，成本平均减少了44%.



Abstract:It is a big challenge to pick up the best cloud configuration for recurring big data analytics jobs running in clouds. Prior efforts may get in a sub-optimal configuration due to a broad spectrum of cloud configurations with a few test runs, such as CherryPick. RP-CH, presented in this paper, is a resource provisioning system that leverages heuristic rules based on classification information to identify the optimal cloud configuration for big data analytics jobs, while the insight is classifying a job by comparing its resource preference and usage information with other jobs. Then, heuristic rules are used to distinguish bad samples from good ones in Bayesian optimization algorithm. The experiments on HiBench and SparkBench in Aliyun ECS show that the performance of job has been improved by 58% in average comparing with CherryPick, meanwhile the resource cost has been reduced by 44% in average.



PDF全文下载地址:

http://jos.org.cn/jos/article/pdf/5710