Hey guys,
I guess Quanwen just misinterpreted the statement by the RWP team as it really
says "Size of connection pool is a single digit factor, times number of CPU
cores available". RWP team is talking about how sizing the connection pool (and
so the parameter PROCESSES and SESSIONS in consequence) but not setting
PROCESSES directly with this formula.
I think the rule of thumb "Size of connection pool is a single digit factor,
times number of CPU cores available" is pretty good (as it is based on an
application think time between 80-90% in layered software architectures). Of
course the best way would be to measure it (e.g. with Extended SQL trace and
profilers like Method-R) but almost nobody puts that time/effort into it for
each application.
Best Regards
Stefan Koehler
Independent Oracle performance consultant and researcher
Website: http://www.soocs.de
Twitter: @OracleSK
Andy Sayer <andysayer@xxxxxxxxx> hat am 11.06.2023 00:35 CEST geschrieben:--
I’m no fan of rules of thumb in general, but 1-3 X CPUs for processes is way
off. Don’t forget the few hundred background processes that you need to
support as standard. The “perfect” number will depend on those (which is
going to be version dependent and have defaults that will be dependent on
other parameters), and the number of foreground processes that your
application realistically requires to achieve throughput without queuing.