Nonstop Database, HiRDB Version 9 System Operation Guide
Reduce the deferred write trigger interval and the number of update buffers during synchronization point processing. If the number of update buffers decreases, the time required for applying the updated pages during synchronization point processing can be reduced, thereby reducing the synchronization point processing time.
The following describes how to tune the deferred write trigger interval.
- Procedure
- Acquire statistical information about the global buffer and the deferred write processing.
- From the execution results in the global buffer statistical information, identify the global buffer whose synchronization point output pages count (SYNCW) is high.
- Use the following operands to specify the deferred write trigger start conditions:
pd_dbbuff_rate_updpage operand
-y option in the pdbuffer operand
Check the global buffer statistical information and change the operand values in such a manner that the following condition is satisfied:
TRGUP (number of update buffers that become output triggers during deferred write triggers) < SYNCW (number of synchronization point dump pages) 2
If the value of TRGUP is too small, the number of write operations increases and the transaction performance might be adversely affected. If this happens, restore the original operand values.
- Check to see if the synchronization point processing time is within 10% of the synchronization point acquisition interval. If it is greater than 10%, repeat steps 1 through 3 until it is within 10%.
To determine the synchronization point processing time, check the execution time (DWTOTAL) in the statistical information about deferred write processing.
Obtain the synchronization point collection interval from the difference between the output times of the following messages:
- KFPS02183-I (message displayed when synchronization point processing is completed)
- KFPS02179-I (message displayed when synchronization point processing is skipped)
Increase the update page output rate during deferred write processing (increase the number of update pages to be applied during deferred write trigger processing), and reduce the number of update buffers during synchronization point processing. If the number of update buffers decreases, the time required for applying the updated pages during synchronization point processing can be reduced, thereby reducing the synchronization point processing time.
The following describes how to tune the update page output rate during deferred write trigger.
- Procedure
- Acquire statistical information about the deferred write processing.
- Increase the -w option value in the pdbuffer operand.
If the -w option value is too large, the number of write operations increases and the transaction performance might be adversely affected. If this happens, restore the original operand value.
- Check to see if the synchronization point processing time is within 10% of the synchronization point acquisition interval. If it is greater than 10%, repeat steps 1 and 2 until it is within 10%.
Obtain the synchronization point collection interval from the difference between the output times of the following messages:
- KFPS02183-I (message displayed when synchronization point processing is completed)
- KFPS02179-I (message displayed when synchronization point processing is skipped)
Increase the number of parallel write processes in deferred write processing to shorten the deferred write processing time. The procedure for tuning the number of parallel write processes in deferred write processing follows.
- Procedure
- Acquire statistical information about the deferred write processing.
- Increase the value of the pd_dfw_awt_process operand to increase the number of parallel write processes in deferred write processing.
Since increasing the value of the pd_dfw_awt_process operand increases the number of parallel write processes in deferred write processing, CPU usage increases. Therefore, use the appropriate OS function or the like to monitor CPU usage.
- Check to see if the synchronization point processing time is within 10% of the synchronization point acquisition interval. If it is greater than 10%, repeat steps 1 and 2 until it is within 10%.
To determine the synchronization point processing time, check the execution time (DWTOTAL) in the statistical information about deferred write processing.
Obtain the synchronization point collection interval from the difference between the output times of the following messages:
- KFPS02183-I (message displayed when synchronization point processing is completed)
- KFPS02179-I (message displayed when synchronization point processing is skipped)
If increasing the number of parallel write processes in deferred write process does nor shorten the parallel WRITE time (DWPARA and DWPARAM), the following are possible explanations:
- Disk contention has occurred.
Check and, if necessary, revise the configuration of disks, RDAREAs, and tables so that input/output processing is not concentrated on any particular volume.
- Global buffer lock-release-wait has occurred.
You can check the global buffer statistical information referred to as lock-release contention rate (SLEPR) in buffer lock processing to determine the occurrence rate of global buffer lock-release wait status. For details about the tuning method, see 23.1.2(11) Check the lock-release contention rate (SLEPR) in global buffer lock processing.
- There are not enough output pages.
The actual maximum number of parallel write processes in deferred write processing can be determined using the following formula:
- Total number of output pages (OUT PAGE)/20
Even if you specify a number that is equal to or greater than this value for the number of parallel write processes in deferred write processing, the number of parallel write processes exceeding this value will not run.
To determine the total number of output pages (OUT PAGE), check statistical information about the deferred write processing.
All Rights Reserved. Copyright (C) 2011, 2015, Hitachi, Ltd.