On 1/2/2013 7:45 PM, Craig Ringer wrote:
> You really need to put a connection pool in place to limit the number of
> concurrent workers. Look at PgBouncer or PgPool-II. As far as I know
> neither of them runs on Windows; you might want to think about a Linux
> box as a front-end.
2nd and 3rd the emotion.
of course, a pooler only works right if the client applications are
modified to open a connection, do a transaction, and close the
connection. if the clients continue to hold idle connections, the
pooler won't do anything useful for you.
typically a database server get the best overall throughput if you limit
the number of concurrent queries to maybe 2-4 times the CPU core count,
give or take how fast your storage IO is and how many pending IO
operations can be in the pipe.
