Optimizing the LOGAPPLY phase

The time that is required to recover a table space depends also on the time that is required to read and apply log data. You can take several steps to optimize the process. If possible, Db2 reads the required log records from the active log to provide the best performance.

Any log records that are not found in the active logs are read from the archive log data sets, which are dynamically allocated to satisfy the requests. The type of storage that is used for archive log data sets is a significant factor in the performance. Consider the following actions to improve performance:

  • RECOVER a list of objects in one utility statement to take only a single pass of the log.
  • Keep archive logs on disk to provide the best possible performance.
  • Control archive logs data sets by using DFSMShsm to provide the next best performance. Db2 optimizes recall of the data sets. After the data set is recalled, Db2 reads it from disk.
  • If the archive log must be read from tape, Db2 optimizes access by means of ready-to-process and look-ahead mount requests. Db2 also permits delaying the deallocation of a tape drive if subsequent RECOVER jobs require the same archive log tape. Those methods are described in more detail in the subsequent paragraphs.

The BSDS contains information about which log data sets to use and where they reside. You must keep the BSDS information current. If the archive log data sets are cataloged, the ICF catalog indicates where to allocate the required data set.

DFSMShsm data sets

The recall of the first DFSMShsm archive log data set starts automatically when the LOGAPPLY phase starts. When the recall is complete and the first log record is read, the recall for the next archive log data set starts. This process is known as look-ahead recalling. Its purpose is to recall the next data set while it reads the preceding one.

When a recall is complete, the data set is available to all RECOVER jobs that require it. Reading proceeds in parallel.

Non-DFSMShsm tape data sets

Db2 reports on the console all tape volumes that are required for the entire job. The report distinguishes two types of volumes:

  • Any volume that is not marked with an asterisk (*) is required for the for the job to complete. Obtain these volumes from the tape library as soon as possible.
  • Any volume that is marked with an asterisk (*) contains data that is also contained in one of the active log data sets. The volume might or might not be required.

As tapes are mounted and read, Db2 makes two types of mount requests:

  • Ready-to-process: The current job needs this tape immediately. As soon as the tape is loaded, Db2 allocates and opens it.
  • Look-ahead: This is the next tape volume that is required by the current job. Responding to this request enables Db2 to allocate and open the data set before it is needed, thus reducing overall elapsed time for the job.

You can dynamically change the maximum number of input tape units that are used to read the archive log by specifying the COUNT option of the SET ARCHIVE command. For example, use the following command to assign 10 tape units to your Db2 subsystem: 

-SET ARCHIVE COUNT (10)

The DISPLAY ARCHIVE READ command shows the currently mounted tape volumes and their statuses.

Delayed deallocation

Db2 can delay deallocating the tape units used to read the archive logs. This is useful when several RECOVER utility statements run in parallel. By delaying deallocation, Db2 can re-read the same volume on the same tape unit for different RECOVER jobs, without taking time to allocate it again.

You can dynamically change the amount of time that Db2 delays deallocation by using the TIME option of the SET ARCHIVE command. For example, to specify a 60 minute delay, issue the following command: 

-SET ARCHIVE TIME(60)

In a data sharing environment, you might want to specify zero (0) to avoid having one member hold onto a data set that another member needs for recovery.

Performance summary

  1. Achieve the best performance by allocating archive logs on disk.
  2. Consider staging cataloged tape data sets to disk before allocation by the log read process.
  3. If the data sets are read from tape, set both the COUNT and the TIME values to the maximum allowable values within the system constraints.