On 13 May 2015 at 09:35, Heikki Linnakangas <hlinnaka@iki.fi> wrote:
In summary, the X^1.5 correction seems to work pretty well. It doesn't completely eliminate the problem, but it makes it a lot better.
Agreed
I don't want to over-compensate for the full-page-write effect either, because there are also applications where that effect isn't so big. For example, an application that performs a lot of updates, but all the updates are on a small number of pages, so the full-page-write storm immediately after checkpoint doesn't last long. A worst case for this patch would be such an application - lots of updates on only a few pages - with a long checkpoint_timeoout but relatively small checkpoint_segments, so that checkpoints are always driven by checkpoint_segments. I'd like to see some benchmarking of that worst case before committing anything like this.
We could do better, but that is not a reason not to commit this, as is. Commit, please.
This has been in place for a while and still remains: TODO: reduce impact of full page writes
--
Simon Riggs http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services