Nonstop Database, HiRDB Version 9 System Operation Guide
Use the pdstbegin command to start statistics log collection. To start collecting statistics logs at the time of HiRDB startup, specify the pdstbegin operand in the system common definition. When the pdstbegin operand is specified, there is no need to execute the pdstbegin command.
- Note
- The information listed below is not collected unless at least two synchronization points occur during statistics log collection. To ensure that this information is collected, use the pdlogsync command to set a synchronization point immediately after entering the pdstbegin command, and then use the pdlogsync command to set a synchronization point immediately before entering the pdstjsync or pdstend command.
- Global buffer pool statistical information
- Statistical information on HiRDB files for data manipulation
- Deferred write processing statistical information
- Index statistical information
- Reference note
- To end statistics log collection, execute the pdstend command.
(2) Outputting statistics logs from the statistics log buffer to a statistics log file
Use the pdstjsync command to output the statistics logs stored in the statistics log buffer to a statistics log file.
- Reference note
- The statistics logs stored in the statistics log buffer are output to a statistics log file at the following trigger points:
- When the statistics log buffer becomes full
- When the pdstjsync command is executed
- When the pdstend command is executed
- When HiRDB is terminated (normal or planned termination)
One of the following methods must be used to unload a statistics log file and create an unload statistics log file; normally, the OS's cp command is used:
- OS's cp command
- HiRDB-provided shell script (pdstjacm)
- Hint
- When statistics log information is being output to multiple server machines (as in the case of a HiRDB parallel server configuration), pdstjacm should be used to create the unload statistics log files. For example, when system activity information is collected by a HiRDB parallel server configuration, the statistics log information is output to multiple server machines. In such a case, pdstjacm must be used to create the unload statistics log files at a specific server machine. For details on pdstjacm, see 22.1.4 Shell script for creating unload statistics log files at a specified server machine.
- Note
- When a statistics log file is unloaded while HiRDB is active, care must be taken not to delete the statistics log file with the OS's mv or rm command. If it is deleted, statistical information will no longer be output. If the statistics log file is deleted inadvertently, either restart HiRDB or execute the pdstjswap command.
(4) Executing the statistics analysis utility
Using the unload statistics log files created in step (3) as the input information, execute the statistics analysis utility. The HiRDB administrator can then use the execution results to tune HiRDB or UAPs.
- Reference note
- Because statistics information is collected between the time when the pdstbegin command is entered and the time when the pdstend (or pdstjsync) command is entered, the UAP statistics information might not match the SQL statistics information depending on the command entry timings.
All Rights Reserved. Copyright (C) 2011, 2015, Hitachi, Ltd.