On Wed, Mar 14, 2018 at 10:58 AM, Claudio Freire <klaussfre...@gmail.com> wrote:
> > > I'm thinking there could be contention on some lock somewhere. > > Can you attach the benchmark script you're using so I can try to reproduce > it? > I am using a very ad-hoc script.. But here it is.. It assumes a presence of a branch named "btree_rightmost" with the patched code. You will need to make necessary adjustments of course. Thanks, Pavan -- Pavan Deolasee http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services
start_options="-c shared_buffers=16GB -c max_wal_size=64GB -c min_wal_size=16GB -c checkpoint_timeout=60min -c bgwriter_delay=100 -c bgwriter_lru_maxpages=100 -c bgwriter_lru_multiplier=3 -c bgwriter_flush_after=256 -c checkpoint_completion_target=0.9" PGDATA=/data/pgsql echo "INSERT INTO testtab(b) SELECT generate_series(1,100000);" > s1.sql for b in master btree_rightmost; do git checkout $b > /dev/null 2>&1 make -s clean > /dev/null 2>&1 ./configure --prefix=$HOME/pg-install/$b make -s -j8 install > /dev/null 2>&1 export PATH=$HOME/pg-install/$b/bin:$PATH for c in 1 2 4 8; do pg_ctl -D $PGDATA -w stop rm -rf $PGDATA initdb -D $PGDATA pg_ctl -D $PGDATA -o "$start_options" -w start -l logfile.$b psql -c "CREATE TABLE testtab (a bigserial UNIQUE, b bigint);" postgres echo "$b $c" >> results.txt num_txns_per_client=$((1024 / $c)) time pgbench -n -l -c $c -j 1 -t $num_txns_per_client -f s1.sql postgres >> results.txt done done