Improve scalability and performance with data partitioning capabilities on Linux

This article first examines the IBM db2®universal database™ data partitioning feature in the DB2 UDB for linux®, UNIX®, and Windows® in the performance and scalability side The advantages of the face. Then, complete the steps to install and configure the DB2 with DPF functionality on the SUSE Linux Enterprise Server. You will also learn some important conceptual and design considerations that will help you quickly master the installation of DPF in the SUSE Linux Enterprise environment.

Brief introduction

When a database is made up of two or more database partitions, and each database partition stores part of the table data in each of the tables, it is the case of the database partition. Different partitions can be on different computers or on the same computer (that is, logical partitions). In this type of database, the data is hashed according to the storage location to identify which database partition will store the data. A database partition consists of its own data, indexes, configuration files, and transaction logic. In older versions of DB2, a database partition is called a node or a database node. But in this article we use the term partition (partition). Logical partitions share resources on the same computer on which they are located.

In a partitioned database, a table can be distributed across one or more database partitions. The idea behind partitioning is to distribute data from one table to multiple partitions, store some of the rows in the table in one partition, and store other rows in other partitions. The Database Manager (DBM) processes the retrieval and updating of data by automatically decomposing SQL requests into child requests. The child requests are then executed in parallel in the corresponding database partition. This behavior of dividing a table into multiple partitions is transparent to the user, which means that although the data is physically stored on multiple database partitions, the access is as if it were in one place.

Figure 1 shows a single partition running on one machine, which is similar to not using DPF:

Figure 1. A single partition on a single machine

The next two graphs show how DPF really works. Figure 2 illustrates the case of DPF being used on three machines, each of which has a partition on it.

Figure 2. One partition per machine, using high-speed communication

Note the high-speed communication interconnection used here, which can be used as a TCPIP network between the machines. The next Figure 3 shows the three partitions on the same machine. Here, the communication between partitions is implemented in memory, not through a network (DB2 _FORCE_FCM_BP is enabled by default on all Linux platforms).

Figure 3. Three partitions on the same machine, communicating through memory