I wrote:
> BTW, according to "top" the CPU usage percentages in these tests are
> on the order of 55% backend, 45% psql. Methinks psql needs a round
> of performance tuning ...
oprofile tells the tale:
CPU: P4 / Xeon with 2 hyper-threads, speed 2793.03 MHz (estimated)
Counted GLOBAL_POWER_EVENTS events (time during which processor is not stopped)
with a unit mask of 0x01 (mandatory) count 240000
GLOBAL_POWER_E...| samples| %|
------------------ 682534 52.7683 /usr/lib/debug/lib/modules/2.6.16-1.2133_FC5/vmlinux 274747 21.2413
/home/tgl/testversion/bin/postgres 226306 17.4962 /lib64/libc-2.4.so 54296 4.1977 /home/tgl/testversion/bin/psql
45376 3.5081 /home/tgl/testversion/lib/libpq.so.5.0 5302 0.4099 /usr/bin/oprofiled 1954 0.1511 /oprofile
It's all about the kernel process-switch overhead, which is being blamed
equally on both processes.
I did find some low-hanging fruit in GetVariable(), but nothing else
that looked readily improvable.
regards, tom lane
