Typically, the synchronization point dump interval is set taking into account the following considerations.

Considerations

• For update processing involving a large amount of data, the transaction processing time is long and there is a large amount of system log information; therefore, recovery at the time of a restart is prolonged.
• For update processing involving a small amount of data, the transaction processing time is short and there is a small amount of system log information; therefore, recovery at the time of a restart is brief.

When the synchronization point dump output interval is tuned, it is also advisable to evaluate the global buffer pool statistical information and the deferred write processing statistical information.

Information to be collected

System activity statistical information is collected by the statistics analysis utility.

Information to be referenced

• Synchronization point dump interval (SYNC POINT GET INTERVAL)
• Synchronization point dump collection time (SYNC POINT GET TIME)

Purpose

In order to determine how appropriate the synchronization point dump interval is, the time required for a restart (time required to collect the system log) and the throughput must be taken into consideration.

The time required for a restart might be shorter than usual, because HiRDB writes updated pages from global buffer to disk during deferred write processing. The performance of synchronization point dumps can be improved if deferred write processing statistical information is also collected and the number of pages written during deferred write processing (OUT PAGE) is evaluated.

Criteria for evaluating the analysis results

1. If many pages are written to disk during a synchronization point dump, the number of pages written to disk during deferred write processing might be too small.
2. If the update buffer hit rate is low and the number of pages output at synchronization points is large, consider reducing the value of the pdbuffer operand's -w option (the rate of pages output in deferred write processing).
3. If the system log file swapping interval is short for the synchronization point dump interval, the size of the system log file might be too small for the number of blocks output at a synchronization point dump interval. This does not apply when the amount of system log file information available for data output is less than the number of blocks to be output at a synchronization point dump. In such a case, a synchronization point dump is collected automatically when the system log file becomes full.

Actions to be taken

1. Set the time required for restart processing in a range acceptable to the user.
2. A synchronization point dump cannot become effective until a transaction that is executing has been completed; therefore, a long transaction should not be executed concurrently with other transactions. A long transaction means that the amount of log information output by one user server during execution of the transaction is greater than one-third the size of the log file for the user server (user server refers to the single server in the case of a HiRDB single server configuration and to a front-end server, back-end server, or dictionary server in the case of a HiRDB parallel server configuration).