Resolving synchronous read I/O problems in data sharing environments
If your data sharing environment experiences a high volume of unexpected synchronous read I/O activity, you can take steps to alleviate the problem.
Symptoms
Causes
- PCLOSET specifies the number of minutes that can elapse after a page set or partition is updated.
- PCLOSEN specifies the number of consecutive checkpoints that are allowed after a page set or partition is updated.
This situation leads to GBP dependency changes, which means that the pages in the local buffer pool are treated as invalid. If those pages are referenced again, they need to be refreshed, perhaps by synchronous read I/O activity. This situation might be worse if the local buffer pool is defined with PGSTEAL(NONE) because this setting disables prefetch.
Environment
Resolving the problem
System programmer response: To resolve this problem, you can take either of the following actions:
- Increase the PCLOSET and PCLOSEN subsystem parameter values to avoid frequent GBP-dependency changes.
- Change the definition of the buffer pool so that it specifies PGSTEAL(LRU), which ensures that the prefetch engines are always enabled. Refreshing invalid pages by the prefetch engines for asynchronous read I/O has less performance impact to applications.
Parent topic: Improving the performance of data sharing applications