Benchmarks comparing Postgres hosting options on GCP using pgbench.

CloudSQL-vs-Crunchy-vs-AlloyDB.md

All tests conducted in us-central1 on GCP from an instance running in us-central1-a

Initialized with:

pgbench -i --foreign-keys -s 100

Tests run with:

pgbench -c 30 -T 120 -s 100 -j 10 -r

Results

CloudSQL: 8 CPU 16GB (Baseline)

scaling factor: 1
query mode: simple
number of clients: 100
number of threads: 1
duration: 120 s
number of transactions actually processed: 43472
latency average = 276.630 ms
tps = 361.493421 (including connections establishing)
tps = 361.539754 (excluding connections establishing)

scaling factor: 1
query mode: simple
number of clients: 30
number of threads: 1
duration: 120 s
number of transactions actually processed: 48894
latency average = 73.678 ms
tps = 407.179195 (including connections establishing)
tps = 407.220811 (excluding connections establishing)

scaling factor: 100
query mode: simple
number of clients: 30
number of threads: 10
duration: 120 s
number of transactions actually processed: 393528
latency average = 9.149 ms
tps = 3278.994506 (including connections establishing)
tps = 3279.593288 (excluding connections establishing)

AlloyDB: 8 CPU 64GB

scaling factor: 1
query mode: simple
number of clients: 100
number of threads: 1
duration: 120 s
number of transactions actually processed: 46500
latency average = 258.595 ms
tps = 386.705171 (including connections establishing)
tps = 386.769029 (excluding connections establishing)

scaling factor: 1
query mode: simple
number of clients: 30
number of threads: 1
duration: 120 s
number of transactions actually processed: 51026
latency average = 70.595 ms
tps = 424.960558 (including connections establishing)
tps = 425.028651 (excluding connections establishing)

scaling factor: 100
query mode: simple
number of clients: 30
number of threads: 10
duration: 120 s
number of transactions actually processed: 469602
latency average = 7.667 ms
tps = 3912.660828 (including connections establishing)
tps = 3914.829176 (excluding connections establishing)

AlloyDB 16 CPU 128GB

scaling factor: 100
query mode: simple
number of clients: 30
number of threads: 10
duration: 120 s
number of transactions actually processed: 736284
latency average = 4.890 ms
tps = 6135.219193 (including connections establishing)
tps = 6137.911798 (excluding connections establishing)

CrunchBridge: 8CPU 32GB

scaling factor: 1
query mode: simple
number of clients: 30
number of threads: 1
duration: 120 s
number of transactions actually processed: 50507
latency average = 71.320 ms
tps = 420.638493 (including connections establishing)
tps = 420.690250 (excluding connections establishing)

scaling factor: 100
query mode: simple
number of clients: 30
number of threads: 10
duration: 120 s
number of transactions actually processed: 236049
latency average = 15.253 ms
tps = 1966.817508 (including connections establishing)
tps = 1967.666898 (excluding connections establishing)

CrunchBridge: 16CPU 64GB

scaling factor: 100
query mode: simple
number of clients: 30
number of threads: 10
duration: 120 s
number of transactions actually processed: 330371
latency average = 10.898 ms
tps = 2752.817835 (including connections establishing)
tps = 2753.973409 (excluding connections establishing)

TPS and Latency (ms).svg