Reduce checkpoint cost in truncateEnd phase of SYSCS_INPLACE_COMPRESS_TABLE
---------------------------------------------------------------------------
Key: DERBY-2912
URL: https://issues.apache.org/jira/browse/DERBY-2912
Project: Derby
Issue Type: Improvement
Components: Performance, Store
Reporter: Daniel John Debrunner
Priority: Minor
The truncate phase of SYSCS_INPLACE_COMPRESS_TABLE always performs a checkpoint
per container. For any table this is likely to mean a number of checkpoints
(one per index and one for the base table).
Potential improvements are:
1) Only perform the checkpoint (and the clean of the allocation cache) if
the truncate will actually free some space up.
2) Perform the checkpoint once for a table, not once per container.
3) Don;t perform a complete checkpoint, just flush the pages for the
table/container being truncated.
1) Seems fairly possible, I have no idea if the others are.
This affects the JUnit tests which perform a inplace compress on SYSDEPENDS,
even with no space to free up the inplace compress takes about 0.5 seconds and
upto two seconds with changes (even on a fairly small table).
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
