功率模型及低功率的查询计划

发布时间：2018-05-29 12:43

本文选题：数据中心 + 能量管理　；参考：《哈尔滨工业大学》2013年硕士论文

【摘要】：随着网络应用的普及和信息采集技术的发展，人类产生和获取数据的能力也迅速发展，导致海量数据几乎无处不在。为了有效支持海量数据的管理和计算，大规模和超大规模数据中心应运而生。但是随着数据中心技术的发展，数据中心消耗的能量越来越多，能量费用占数据中心成本的比例也逐年上升。因此，能量消耗已经成为数据中心中不容忽视的问题，能量管理技术也越来越重要，已经成为数据中心的一个重要的度量标准和设计因素。如何准确预测系统消耗的功率是进行有效能量管理的基础，而预测系统消耗的功率主要通过构建功率模型来完成。但是，，现有的功率模型只是针对单核CPU，而没有考虑多核CPU的情况，而且随着硬件技术的发展多核CPU服务器越来越普遍。因此，本文首先基于多核CPU提出了多核CPU功率模型，其次在考虑CPU利用率的情况下提出了一个新的内存功率模型，最后根据本文提出的多核CPU功率模型、内存功率模型以及磁盘功率模型构建了一个新的整机功率模型。真机实验结果表明本文提出的多核CPU功率模型和内存功率模型比现有的功率模型在预测功率消耗方面更精确。 DBMS作为数据中心中一个重要软件，承担了数据中心中的大部分任务负载，同时也消耗了数据中心中大部分能量。但现在很少有研究关注DBMS查询执行计划的功率有效性。现有的DBMS生成的查询执行计划只考虑了时间代价，而没有考虑功率代价。针对这种情况，本文考虑重新设计查询优化器，生成功率有效的查询执行计划。本文构建了查询计划功率模型来预测查询计划基础操作的功率消耗，提出了生成功率有效的查询执行计划的算法。通过基于TPC-H标准的实验，证明了查询计划功率模型的准确性，生成了功率有效的查询执行计划。本文比较了传统的查询执行计划与功率有效的查询执行计划，证明了算法的有效性。
[Abstract]:With the popularization of network applications and the development of information collection technology, the ability of human to produce and obtain data has developed rapidly, resulting in massive data almost everywhere. In order to effectively support the management and computation of massive data, large-scale and large-scale data centers emerge as the times require. However, with the development of data center technology, more and more energy is consumed by data center, and the proportion of energy cost to data center cost is increasing year by year. Therefore, energy consumption has become a problem that can not be ignored in the data center, and energy management technology is becoming more and more important, which has become an important metric and design factor in the data center. How to accurately predict the power consumed by the system is the basis of effective energy management, and the power consumption of the system is mainly predicted by building a power model. However, the existing power model is only for single core CPU, and does not consider the case of multi-core CPU, and with the development of hardware technology, multi-core CPU server is becoming more and more common. Therefore, this paper first proposes a multi-core CPU power model based on multi-core CPU, then proposes a new memory power model considering the CPU utilization, finally, according to the multi-core CPU power model proposed in this paper, The memory power model and the disk power model are used to construct a new power model. The experimental results show that the proposed multi-core CPU power model and the memory power model are more accurate than the existing power model in predicting power consumption. As an important software in the data center, DBMS takes on most of the task load in the data center and consumes most of the energy in the data center. However, little research has focused on the power efficiency of DBMS query execution plan. The existing query execution plan generated by DBMS only considers the time cost, not the power cost. In view of this situation, this paper considers redesigning query optimizer to produce effective query execution plan. In this paper, a query plan power model is constructed to predict the power consumption of the basic operation of the query plan, and an efficient query execution plan algorithm is proposed. Through the experiment based on TPC-H standard, the accuracy of the query plan power model is proved, and the efficient query execution plan is generated. This paper compares the traditional query execution plan with the power efficient query execution plan, and proves the effectiveness of the algorithm.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP308

【相似文献】