Simon,
> Not checkpointing at all is not a good plan, since this will lead to an
> enormous build up of WAL files and a very long recovery time if the
> system does fail.
I appreciate your detailed comments.
Following your comments, I revised the problem.
Sigres-0.1.3 does checkpointings.
In summary, the features of Sigres-0.1.3 are as follows.
0: 10% faster than conventional PostgreSQL under tmpfs.
1: Checkpointings are continually executed.
2: Sigres mode is in default (the mode can be turned off via postgresql.conf).
3: issue_xlog_sync is called only by bgwriter (continually, via
createcheckpoint)
4: The entity of XLogWrite (_XLogWrite in my code) is called by both
backends and a bgwriter.
For each backend, _XLogWrite is called only via AdvanceXLInsertBuffer.
For a bgwriter, _XLogWrite is called via CreateCheckPoint.
Please try it if you have interest.
http://sourceforge.jp/projects/sigres/
Again, I really appreciate beneficial comments from this community !
Regards,
-- Hideyuki
--
Hideyuki Kawashima (Ph.D.)
University of Tsukuba
Assistant Professor