Relational database management systems currently keep large volumes of data in secondary storage. But with the increasing development of hardware features such as quantity of memory, number of CPUs and CPUs cores, computation cycles, etc. the new trend is to store data in main memory. Such storage will lead to a suited different organization of data that only is efficient if data is always available in main memory. Taking into account this idea, SAP has developed a new relational DBMS: SAP HANA. It is an implementation of in-memory database technology designed for enterprise computing. In this database it is necessary to separate data into two differenced categories: cold data and hot data. This separation is necessary to handle the increasing, but still limited capacity of main memory efficiently while keeping all data available for reporting needs. Data in a database has its own life cycle and the sequence of events that certain data has passed through determines its state. The term cold data refers to data that has become passive and can be stored in disk if it will not be changed any longer and, thus, this data will be accessed less often. In the other hand, the term hot data is used to refer data that is frequently accessed. The disadvantage with in-memory databases is the memory consumption especially for data which are rarely accessed, i.e. cold data. Therefore some mechanisms of streaming cold data in/out disk without using any kind of buffering, which allows consuming the minimal amount of memory, have been designed, implemented and evaluated. According to experimental results, the approach that consists of using a column store data model and applying some compression techniques show the best performance. ___________________________________________________________________________________________________________________________________
展开▼
