Consider a throttled read-only load composed of 3 random selects,
several rounds (so.sql) vs one (so2.sql):
> pgbench -h server -T 100 -P 1 --rate 200 -c 2 -f so.sql -f so2.sql
SQL script 1: so.sql
- latency average = 1.878 ms
SQL script 2: so2.sql
- latency average = 1.089 ms
There is 0.8 ms latency reduction, that is a 40% difference. Standard
deviation is also significantly lower.
More fun with these two scripts, for those who focus on tps.
> pgbench -h server -M prepared -T 10 -P 1 -c 8 -f so.sql
...
latency average = 1.128 ms
latency stddev = 0.167 ms
tps = 7024.818252 (including connections establishing)
> pgbench -h server -T 10 -P 1 -c 8 -f so2.sql
# cannot do "-M prepared" on multiple commands... anyway:
...
latency average = 0.454 ms
latency stddev = 0.110 ms
tps = 17317.807554 (including connections establishing)
A mere 2.46 speedup for the read-only load.
The same with a "simple update" pgbench load, not compound & prepared:
latency average = 1.716 ms
latency stddev = 1.890 ms
tps = 4623.270608 (including connections establishing)
vs compound but not prepared:
latency average = 1.294 ms
latency stddev = 0.870 ms
tps = 6122.289249 (including connections establishing)
Ok, it is less impressive, just 30% better.
--
Fabien.
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers
