pgbench-results.txt

# BENCHMARK

A single backend does `LISTEN mychannel;` and stays idle,
then pgbench is run 3 times for each <N backends>.

script.sql: NOTIFY mychannel;

% pgbench" -f script.sql -c <N backends> -j <N backends> -T 10 -n

## TPS

 N backends | master | patch-v1 | patch-v2
------------+--------+----------+----------
          1 | 117343 |   151422 |   150735
          2 | 158427 |   236705 |   239004
          4 | 177454 |   250783 |   250782
          8 | 116521 |   155466 |   180418
         16 |  45627 |   144740 |   163491
         32 |  37281 |   135602 |   146659
         64 |  36608 |   123870 |   131202
        128 |  34798 |   120302 |   119041
(8 rows)


## TPS speed-up vs master

 N backends | patch-v1 | patch-v2
------------+----------+----------
          1 | +29%     | +28%
          2 | +49%     | +51%
          4 | +41%     | +41%
          8 | +33%     | +55%
         16 | +217%    | +258%
         32 | +264%    | +293%
         64 | +238%    | +258%
        128 | +246%    | +242%
(8 rows)