PostgreSQL Benchmarks with pgbench and pg_test_fsync on an Intel Optane 900P 480GB PCIe SSD (w/ Ryzen 3600X amd X570 chipset)

pg_test_fsync

$ uname -a
Linux owl 5.4.8-arch1-1 #1 SMP PREEMPT Sat, 04 Jan 2020 23:46:18 +0000 x86_64 GNU/Linux
$ cat /etc/issue 
Arch Linux \r (\l)
$ pg_config --version
PostgreSQL 12.1
$ cd /var/lib/postgres/data/

$ pg_test_fsync
5 seconds per test
O_DIRECT supported on this platform for open_datasync and open_sync.

Compare file sync methods using one 8kB write:
(in wal_sync_method preference order, except fdatasync is Linux's default)
        open_datasync                     33125.867 ops/sec      30 usecs/op
        fdatasync                         32820.947 ops/sec      30 usecs/op
        fsync                             30742.545 ops/sec      33 usecs/op
        fsync_writethrough                              n/a
        open_sync                         31870.873 ops/sec      31 usecs/op

Compare file sync methods using two 8kB writes:
(in wal_sync_method preference order, except fdatasync is Linux's default)
        open_datasync                     27483.496 ops/sec      36 usecs/op
        fdatasync                         28573.229 ops/sec      35 usecs/op
        fsync                             28281.408 ops/sec      35 usecs/op
        fsync_writethrough                              n/a
        open_sync                         27087.070 ops/sec      37 usecs/op

Compare open_sync with different write sizes:
(This is designed to compare the cost of writing 16kB in different write
open_sync sizes.)
         1 * 16kB open_sync write         28103.020 ops/sec      36 usecs/op
         2 *  8kB open_sync writes        27073.324 ops/sec      37 usecs/op
         4 *  4kB open_sync writes        15794.683 ops/sec      63 usecs/op
         8 *  2kB open_sync writes         4265.749 ops/sec     234 usecs/op
        16 *  1kB open_sync writes         1818.145 ops/sec     550 usecs/op

Test if fsync on non-write file descriptor is honored:
(If the times are similar, fsync() can sync data written on a different
descriptor.)
        write, fsync, close               31816.711 ops/sec      31 usecs/op
        write, close, fsync               31824.815 ops/sec      31 usecs/op

Non-sync'ed 8kB writes:
        write                            590913.573 ops/sec       2 usecs/op

pgbench

$ uname -a
Linux owl 5.4.8-arch1-1 #1 SMP PREEMPT Sat, 04 Jan 2020 23:46:18 +0000 x86_64 GNU/Linux
$ cat /etc/issue 
Arch Linux \r (\l)
$ pg_config --version
PostgreSQL 12.1

$ pgbench -i -h localhost -p 5432 -U postgres -s 25

$ pgbench -h localhost -p 5432 -U postgres -t 1000
starting vacuum...end.
transaction type: <builtin: TPC-B (sort of)>
scaling factor: 25
query mode: simple
number of clients: 1
number of threads: 1
number of transactions per client: 1000
number of transactions actually processed: 1000/1000
latency average = 0.314 ms
tps = 3187.503741 (including connections establishing)
tps = 3199.668095 (excluding connections establishing)

[postgres@owl data]$ pgbench -h localhost -p 5432 -U postgres -t 10000 -c 90 -t 20
starting vacuum...end.
transaction type: <builtin: TPC-B (sort of)>
scaling factor: 25
query mode: simple
number of clients: 90
number of threads: 1
number of transactions per client: 20
number of transactions actually processed: 1800/1800
latency average = 10.575 ms
tps = 8510.254512 (including connections establishing)
tps = 8545.540015 (excluding connections establishing)