Re: Some performance degradation in REL_16 vs REL_15

On 30.10.2023 22:51, Andres Freund wrote:

> There's really no point in comparing peformance with assertions enabled
> (leaving aside assertions that cause extreme performance difference, making
> development harder). We very well might have added assertions making things
> more expensive, without affecting performance in optimized/non-assert builds.
>

Thanks for advice! I repeated measurements on my pc without asserts and CFLAGS="-O2".
Also i reduced the number of clients to -c6 to leave a reserve of two cores
from my 8-core cpu and used -j6 accordingly.

The results were similar: on my pc REL_10_STABLE(c18c12c9) was faster than REL_16_STABLE(07494a0d)
but the effect became weaker:
REL_10_STABLE gives ~965+-15 TPS(+-2%) while REL_16_STABLE gives ~920+-30 TPS(+-3%) in the test: pgbench -s8 -c6 -T20 -j6
So 10 is faster than 16 by ~5%. (see raw-my-pc.txt attached for the raw data)

Then, thanks to my colleagues, i carried out similar measurements on the more powerful 24-core standalone server.
The REL_10_STABLE gives 8260+-100 TPS(+-1%) while REL_16_STABLE gives 8580+-90 TPS(+-1%) in the same test: pgbench -s8 -c6 -T20 -j6
The test gave an opposite result!
On that server the 16 is faster than 10 by ~4%.

When i scaled the test on server to get the same reserve of two cores, the results became like this:
REL_10_STABLE gives ~16000+-300 TPS(+-2%) while REL_16_STABLE gives ~18500+-200 TPS(+-1%) in the scaled test: pgbench -s24 -c22 -T20 -j22
Here the difference is more noticeable: 16 is faster than 10 by ~15%. (raw-server.txt)

The configure options and test scripts on my pc and server were the same:
export CFLAGS="-O2"
./configure --enable-debug --with-perl --with-icu --enable-depend --enable-tap-tests
#reinstall
#reinitdb
#create database bench
for ((i=0; i<100; i++)); do pgbench -U postgres -i -s8 bench> /dev/null 2>&1;
psql -U postgres -d bench -c "checkpoint"; RES=$(pgbench -U postgres -c6 -T20 -j6 bench;

Configurations:
my pc: 8-core AMD Ryzen 7 4700U @ 1.4GHz, 64GB RAM, NVMe M.2 SSD drive.
Linux 5.15.0-88-generic #98~20.04.1-Ubuntu SMP Mon Oct 9 16:43:45 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux
server: 2x 12-hyperthreaded cores Intel(R) Xeon(R) CPU X5675 @ 3.07GHz, 24GB RAM, RAID from SSD drives.
Linux 5.10.0-21-amd64 #1 SMP Debian 5.10.162-1 (2023-01-21) x86_64 GNU/Linux

I can't understand why i get the opposite results on my pc and on the server. It is clear that the absolute
TPS values will be different for various configurations. This is normal. But differences?
Is it unlikely that some kind of reference configuration is needed to accurately
measure the difference in performance. Probably something wrong with my pc, but now
i can not figure out what's wrong.

Would be very grateful for any advice or comments to clarify this problem.

With the best wishes!

--
Anton A. Melnikov
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company