Nonstop Database, HiRDB Version 9 System Operation Guide
When a large amount of data is updated, the number of update buffers that must be applied to the database during synchronization point processing (number of update pages in the global buffer) increases, resulting in a corresponding increase in the time required for synchronization point processing. If synchronization point processing is skipped two or more times in succession, the number of system log files that cannot be overwritten increases. If there is no swappable file, HiRDB (or the unit in the case of a HiRDB parallel server configuration) terminates abnormally.
To avoid this, it is important to complete synchronization point processing within the amount of time that was estimated at the design stage. If the following conditional expression is satisfied, there is no need to perform tuning:
If synchronization point processing is skipped due to a delay in deferred write processing, the KFPS02179-I message (factor code = A01-01) is displayed. In such a case, perform tuning according to the procedure described below.
Acquire the following tuning information:
To acquire 1, use the appropriate OS function. To acquire 2 and 3, use the statistics analysis utility.
Check the statistical information about the deferred write processing whose cause (CAUSE) is S (synchronization point processing). If one of the following conditions is satisfied, proceed to (3). If neither of the conditions is satisfied, proceed to (9).
Obtain the synchronization point acquisition interval from the difference between the output times of the following messages:
If the average WRITE unit time for the statistical information about deferred write processing (DWAVG and DWAVGM) is extremely poor compared to the disk write performance, do the following (if these do not apply, proceed to (4)):
If synchronization point processing is skipped, the KFPS02179-I message is displayed. If any of the utilities listed below was executing during the period in which this message was displayed, consider changing the operating method (if this does not apply, proceed to (5)):
Evaluate whether the synchronization point acquisition interval can be made longer. If it cannot be made longer, proceed to (6). You use the pd_log_sdinterval operand to change the synchronization point acquisition interval.
Perform the tuning described in 23.3.3(1) Reducing the deferred write trigger interval and in 23.3.3(2) Increasing the update page output rate during deferred write trigger. If the problem is not resolved after tuning, proceed to (7).
If there is enough room in the CPU, perform the tuning described below. If there is not enough room in the CPU, proceed to (8).
Specify the pd_dfw_syncpoint_skip_limit operand to limit the number of update buffers. In this case, note that when the skip count for synchronization point processing reaches the maximum value, the performance of an update transaction decreases because the update buffer is output at the extension of the update transaction.
If the pd_dfw_syncpoint_skip_limit operand cannot be specified, evaluate a disk enhancement.
If any of the following events has occurred, tune the global buffer:
For details about the above information and the tuning of the global buffer, see 23.1 Tuning global buffer pools.
All Rights Reserved. Copyright (C) 2011, 2015, Hitachi, Ltd.