With this method, reference buffers and update buffers are managed by independent LRUs.
If a shortage occurs in the global buffers, the least recently accessed reference buffer from among the global buffers is removed from the memory.
(a) Criteria
In the following case, it is preferable for the reference buffer and the update buffer to be managed in separate LRUs:
- There is a relatively small amount of update processing compared to retrieval processing, and the update buffer hit rate is high (the number of reference or update operations per transaction is relatively small, such as in the case of online applications).
(b) Specification
SEPARATE is specified in pd_dbbuff_lru_option in the system common definition.
(c) Notes
- If a large amount of update processing occurs, the reference buffer hit rate drops, slowing down retrieval processing.
- If either of the following applies, MIX is unconditionally assumed for the pd_dbbuff_lru_option operand. For this reason, the reference and update buffers cannot be managed by independent LRUs.
- commit is specified in the pd_dbsync_point operand
- N is specified in the pd_dbbuff_binary_data_lru operand
With this method, the global buffers are managed collectively by one LRU.
If a shortage occurs in the global buffers, the least recently accessed buffer from among the global buffers is removed from the memory.
(a) Criteria
It is beneficial to employ batch LRU management of global buffers in the following case:
- There is more update processing than retrieval processing, or a large amount of data is retrieved or updated sporadically (both retrieval processing and update processing involving a large amount of data occur, such as when online applications and batch applications co-exist).
(b) Specification
MIX is specified in pd_dbbuff_lru_option in the system common definition.
The updated output page rate for deferred write trigger is specified in the -w option of the pdbuffer operand in the system common definition.
(c) Notes
- If the update buffer hit rate is high, update buffers may be removed temporarily from memory due to retrieval of a large amount of data. In such a case, file read operations may occur as an extension of update processing, slowing down the processing.
- If pd_dbsync_point=sync is specified or omitted, file write operations may occur as an extension of retrieval processing, slowing down the retrieval processing.