Research On Energy-Saving Query Processing Technologies For Database Clusters

With the advent of big data era, the data centers need to deal with an increasing amount of data, hence the number and scale of global data centers are in rapid expansion. The expansion of Data centers has brought higher processing performance at the same time, also caused a number of problems, among which the huge energy consumption is the most serious and cannot be ignored. Firstly, the enterprise must pay a huge fee of energy consumption every year to support the data center operation. Secondly, in order to meet the highest performance requirement, data centers tend to configure a large amount of server clusters, which leads to an obvious phenomenon of energy waste, since most of the servers in cluster are idle or underutilized. Finally, a huge energy consumption will cause some environmental issues, such as the rapid consumption of energy sources and carbon emissions. Therefore, no matter from the angle of the enterprise or the perspective of environmental protection, it is necessary to put forward some effective energy consumption management basing on the cluster environment.As the most common system in the data center, the energy consumption management issues of database clusters needs to be discussed thoroughly. With the global trend of low carbonization and the trend of data centric computation, studying an energy saving database system has become a common concern of government, enterprises and academia. This paper mainly studies the energy consumption management of the database cluster environment. Since the query operation is the most common operation in DBMS, reconsidering the query processing to put forward corresponding energy-saving technology can effectively reduce the energy consumption. By now, the researches in the area of cluster energy management can be divided into two direction:one is the energy proportionality and the other is energy efficiency. The pursuit and realization method of these two methods are distinct. However, the researches in the area of database cluster are still relatively short, the existing algorithms transplanted directly into the database cluster will significantly weaken the query processing capabilities of cluster. So in this paper we proposed the methods to realize the energy proportionality and efficiency of a database cluster in the premise of ensuring the query performance.Specifically, the main contribution of this paper can be summarized as follows:(1) As for the research on energy proportionality, we introduced a new architecture for energy proportional database cluster.it is a hybrid architecture combining the share-nothing and share-disk architectures. Furthermore, we introduced a query stream buffer on top of the hybrid architecture to cache extra queries, so that the cluster can still keep high performance even when it is overloaded. Based on the architecture we proposed the unbalanced load allocation algorithm to distribute the workload among the nodes in the cluster to realize better energy proportionality. At last, we presented new algorithm for node activation/deactivation to switch the state of servers. Compared to the existing algorithms, our algorithm can dramatically reduce the erroneous judgment phenomenon and save more energy.(2) AS for the research on energy efficiency, we adapted the DVFS algorithm of single DBMS into database cluster, overcoming the deficiencies of single DVFS that it behaves poor performance when executing the concurrent queries. We divided the servers of cluster into a few sets with different CPU frequency, and queries will also be divided according to their characteristics. In this way we can allocate different kind of queries to the nodes with distinct CPU frequency to achieve the goal of energy efficiency and high performance. Finally, we conducted the experiments in PostgreSQL cluster and validity of this method are proved.