On 9/28/2013 11:29 AM, Tomas Vondra wrote:
> There are probably some corner cases where this might improve the
> performance, but in most cases it's going to be worse. Why are you
> switching to multiple clusters?
>
> For example consider that you'll probably have to use much smaller shared
> buffers (which might cause issues on the active database, while the other
> databases don't use their portion of memory), you'll have to either use
> much smaller max_connections or decrease work_mem (you can't just keep the
> values because then you might run into OOM much more frequently) etc.
indeed, 10 separate checkpoint processes going off asynchronously, meh.
--
john r pierce 37N 122W
somewhere on the middle of the left coast