基于SSD数据库负载的SQL能耗感知模型

SQL energy consumption perception model for database load based on SSD

面对大数据带来的能耗及环境方面的严峻问题,构建节能的绿色数据库系统已成为关键需求和重要挑战。针对现有数据库系统主要以性能优化为目标,缺少对能耗的感知及优化的问题,提出基于数据库负载的能耗感知模型,并将模型应用于基于固态硬盘(SSD)的数据库系统中。首先,将数据库负载执行过程中对主要系统资源(CPU、固态硬盘)的消耗解析为时间开销和功耗开销,并基于SSD数据库负载的基本I/O类型构建时间开销模型和功耗开销模型,实现为数据库构建资源开销单位统一的能耗感知模型;然后,利用多元线性回归实现对模型的求解,并分别在独占环境和竞争环境下,验证模型对不同I/O类型的数据库负载能耗估算的准确性;最后,分析实验结果,并讨论了影响模型准确性的因素。经实验验证模型准确度较高,在DBMS独占系统资源情况下的平均误差为5. 15%,绝对误差不超过9. 8%;竞争环境下的准确率相对下降,但平均误差也低于12. 21%,可有效构建能耗感知的绿色数据库系统。

For energy consumption and severe environmental problems brought by big data, building an energy-efficient green database system has become a key requirement and an important challenge. To solve the problem that traditional database systems mainly focus on performance, and are lack of energy consumption perception and optimization, an energy consumption perception model based on database workload was proposed and applied to the database system based on SolidState Drive( SSD). Firstly, the consumption of major system resources( CPU, SSD) during database workload execution was quantified as time overhead and power consumption overhead. Based on basic I/O type of SSD database workload, a time cost model and a power consumption overhead model were built, and an energy consumption perception model with uniform resource unit was implemented. Then, multi-variable linear regression mathematical tools were used to solve the model, and in the exclusive environment and competitive environment, the energy estimation accuracy of the model for different I/O types of database workload was verified. Finally, the experimental results were analyzed and the factors that affect the model accuracy were discussed. The experimental results show that the model accuracy is relatively high. Under ideal conditions that DBMS monopolized system resources, the average error is 5. 15% and the absolute error is no more than 9. 8%. Although the accuracy in competitive environment is reduced, the average error is less than 12. 21%. The model can effectively build an energy-aware green database system.