DataPower comparison

PostgreSQL, Neo4j Community, TETRA, lance-csr and SQLite

AMD EPYC 7502P 32-Core Processor, 64 cores, 252 GiB, kernel 6.12.67 · dpbench aec78a9

Each system ran the same scenarios against the same generated data on the same host. A system keeps its colour on every chart: PostgreSQL, Neo4j Community, TETRA, lance-csr, SQLite.

The standard workload, envelope S, 2 cores and 8 GiB, blended 80 / 15 / 5

15 runs at SF-300K, SF-1M, SF-3M.

Results

SystemScaleDPDP-ReadDP-WriteET ops/sk*ROM GiBDisk GiBBytes per elementCPU ms per opVerdict
PostgreSQL 16.15SF-300K 98105130 0.301,152.342.0821.70 0.630.642,2901.59Non-standard
PostgreSQL 16.15SF-1M 172167373 1.00601.742.0581.70 0.440.555883.05Non-standard
PostgreSQL 16.15SF-3M 198140683 3.00230.142.0561.70 0.410.913257.54Non-standard
Neo4j Community 5.26SF-300K 3339153 0.30261.442.1571.10 1.230.592,1206.11Non-standard
Neo4j Community 5.26SF-1M 5558450 1.00130.642.1711.10 1.321.321,41610.96Non-standard
Neo4j Community 5.26SF-3M 67681,254 3.0054.942.2281.10 1.673.921,40225.89Non-standard
TETRA v0SF-300K 8034620 0.30552.612.0681.00 0.540.041581.53Non-standard
TETRA v0SF-1M 26573067 1.00557.642.1071.00 0.850.121252.12Non-standard
TETRA v0SF-3M 5201,000193 3.00375.842.1671.00 1.330.23822.94Non-standard
lance-csr 0.5.1SF-300K 6133 0.3058.812.0711.35 0.560.026916.52Non-standard
lance-csr 0.5.1SF-1M 122612 1.0035.712.1561.35 1.250.044020.57Non-standard
lance-csr 0.5.1SF-3M 26365 3.0027.712.4141.35 3.310.113927.89Non-standard
SQLite 3.53.4SF-300K 7417355 0.30742.812.0111.50 0.090.072500.99Non-standard
SQLite 3.53.4SF-1M 109205105 1.00329.612.0121.50 0.090.09932.58Non-standard
SQLite 3.53.4SF-3M 112178176 3.00113.142.0151.50 0.110.19669.09Non-standard

Scaling

DataPower DP at each scale. A rejected run has no DataPower and leaves a gap.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
1310301003001000DP300K1M3MTETRAPostgreSQLSQLiteNeo4j Communitylance-csr
Useful throughput T Operations per second at k*, discounted by Q for every class over its p99 target.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
10301003001000300010kops/s300K1M3MTETRAPostgreSQLSQLiteNeo4j Communitylance-csr
Resources R Cores, plus the memory charge and the disk footprint priced in resource units.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
0123RU300K1M3Mlance-csrNeo4j CommunityTETRAPostgreSQLSQLite
Memory charge M Time-weighted memory.current over the duty cycle at k*.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
01234GiB300K1M3Mlance-csrNeo4j CommunityTETRAPostgreSQLSQLite
Disk per element Bytes on the data volume for each data element, at rest.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
0100020003000bytes300K1M3MNeo4j CommunityPostgreSQLTETRASQLitelance-csr
CPU per operation CPU milliseconds the SUT spent on each standard operation.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
0102030ms300K1M3Mlance-csrNeo4j CommunitySQLitePostgreSQLTETRA

Concurrency

Concurrency ladder at SF-300K Useful throughput X × Q at each rung of the mixed ladder. The highest point sets T, and its rung is k*.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
050010001500ops/s141664PostgreSQLSQLiteTETRANeo4j Communitylance-csr
Concurrency ladder at SF-1M Useful throughput X × Q at each rung of the mixed ladder. The highest point sets T, and its rung is k*.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
0200400600800ops/s141664PostgreSQLSQLiteTETRANeo4j Communitylance-csr
Concurrency ladder at SF-3M Useful throughput X × Q at each rung of the mixed ladder. The highest point sets T, and its rung is k*.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
0100200300400ops/s141664TETRAPostgreSQLSQLiteNeo4j Communitylance-csr

From the cold restart

Each line starts when the run restarts the SUT with an empty page cache (P3), and follows the warm-up (P4) and the warm single-client run (P5). The scored phases come later and start warm, so these charts carry the part of each system's behaviour the score leaves out today.

SF-300K

Throughput from the cold restart Valid operations per second: one client in P3, the warm-up's clients in P4, one client again in P5.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
0500100015002000ops/s0:000:100:200:300:400:50TETRAPostgreSQLNeo4j Communitylance-csrSQLite
Disk reads from the cold restart MB/s the SUT read from disk while its caches filled.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
02505007501000MB/s0:000:100:200:300:400:50PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
Memory from the cold restart memory.current of the SUT's cgroup, the figure M charges.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
00.511.5GiB0:000:100:200:300:400:50Neo4j Communitylance-csrTETRAPostgreSQLSQLite
Page cache from the cold restart File pages the kernel holds for the SUT and can reclaim.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
00.20.40.6GiB0:000:100:200:300:400:50lance-csrNeo4j CommunityPostgreSQLTETRASQLite

SF-1M

Throughput from the cold restart Valid operations per second: one client in P3, the warm-up's clients in P4, one client again in P5.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
050010001500ops/s0:000:100:200:300:400:501:00PostgreSQLSQLitelance-csrNeo4j CommunityTETRA
Disk reads from the cold restart MB/s the SUT read from disk while its caches filled.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
050100150MB/s0:000:100:200:300:400:501:00PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
Memory from the cold restart memory.current of the SUT's cgroup, the figure M charges.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
00.511.5GiB0:000:100:200:300:400:501:00Neo4j Communitylance-csrTETRAPostgreSQLSQLite
Page cache from the cold restart File pages the kernel holds for the SUT and can reclaim.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
00.20.40.6GiB0:000:100:200:300:400:501:00lance-csrNeo4j CommunityPostgreSQLTETRASQLite

SF-3M

Throughput from the cold restart Valid operations per second: one client in P3, the warm-up's clients in P4, one client again in P5.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
02505007501000ops/s0:000:301:00TETRAPostgreSQLlance-csrNeo4j CommunitySQLite
Disk reads from the cold restart MB/s the SUT read from disk while its caches filled.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
02505007501000MB/s0:000:301:00Neo4j CommunityPostgreSQLTETRAlance-csrSQLite
Memory from the cold restart memory.current of the SUT's cgroup, the figure M charges.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
0123GiB0:000:301:00lance-csrNeo4j CommunityTETRAPostgreSQLSQLite
Page cache from the cold restart File pages the kernel holds for the SUT and can reclaim.
PostgreSQLNeo4j CommunityTETRAlance-csrSQLite
00.20.40.60.8GiB0:000:301:00Neo4j Communitylance-csrPostgreSQLSQLiteTETRA

Latency against target

p99 at k* in the mixed ladder, in milliseconds. Blue cells beat the target, grey cells sit on it, and red cells miss it; the colour deepens with each factor of ten.

ClassTargetPostgreSQL
SF-300K
PostgreSQL
SF-1M
PostgreSQL
SF-3M
Neo4j Community
SF-300K
Neo4j Community
SF-1M
Neo4j Community
SF-3M
TETRA
SF-300K
TETRA
SF-1M
TETRA
SF-3M
lance-csr
SF-300K
lance-csr
SF-1M
lance-csr
SF-3M
SQLite
SF-300K
SQLite
SF-1M
SQLite
SF-3M
R1 point106.39.89.714.012.312.60.617.331.616.318.418.60.20.310.0
R1w point, worst case107.28.28.98.110.49.30.611.120.416.924.815.00.20.35.9
R2 filter2006.612.14.619.711.711.52.417.031.913.617.826.80.40.510.7
R2w filter, worst case2008.110.911.333.737.297.65.532.662.88.914.034.02.04.310.4
R3 neighbourhood506.67.69.312.615.59.90.714.119.829.529.233.70.30.410.2
R3w neighbourhood, worst case505.317.3110.012.324.352.31.615.529.935.236.662.21.73.724.4
R4 deep5006.59.09.213.918.218.20.918.315.922.827.624.60.70.88.4
R4w deep, worst case5003.24.56.010.015.014.21.620.923.931.830.051.01.41.315.5
R5 path5008.211.1100.810.613.011.60.626.722.416.120.016.84.413.188.5
R5w path, worst case5004.912.915.711.81.811.30.68.051.16.17.111.63.07.465.2
R6 pattern2,0004.013.04.26.713.26.910.015.325.036.538.546.20.30.44.7
R6w pattern, worst case2,0001.50.40.415.412.116.94.812.437.335.638.348.40.20.31.2
R7 analytic5,00054.8146.6400.1118.4209.2697.914.554.2115.722.539.671.118.853.9309.5
R7w analytic, worst case5,00041.2138.2440.3107.8250.4574.515.436.292.118.632.670.517.950.7266.0
S1 amount band1,00083.5104.3183.2266.0622.11,593.310.822.331.217.551.2102.311.620.5131.8
S3 text search2,00013.751.7126.445.5113.1303.91.019.215.511.218.949.04.614.8103.7
J4 neighbours alike50011.923.672.115.210.110.31.110.219.228.140.043.13.812.7106.2
O1 person by email2012.325.369.128.088.3148.60.719.115.815.621.026.31.44.436.6
O2 accounts by opening date2009.418.057.646.1136.1257.50.929.132.210.811.511.01.424.4321.3
O3 people by surname and city10011.025.661.311.59.316.21.116.116.022.927.237.31.54.538.4
W1 insert2014.217.912.221.529.623.44.332.129.827.432.6230.02.42.4133.8
W2 insert edge1013.214.215.720.328.837.23.928.826.710.558.98.01.71.7133.8
W3 batch50071.469.185.6122.8194.0124.428.568.258.512.513.912.833.534.351.1
U1 modify209.913.48.117.727.811.11.828.619.838.956.472.61.61.6105.0
U2 rewire3012.719.321.341.022.131.48.730.523.830.816.015.61.92.439.2
D1 delete edge208.910.911.823.858.416.16.615.815.845.927.249.51.51.918.9
D2 delete cascade10026.962.1191.915.612.937.32.310.667.8108.4164.2165.622.137.4221.8

Deviations

Every run here carries these:

PostgreSQL at SF-300K also carries:

PostgreSQL at SF-3M also carries:

Neo4j Community at SF-300K also carries:

Neo4j Community at SF-1M also carries:

Neo4j Community at SF-3M also carries:

TETRA at SF-300K also carries:

TETRA at SF-3M also carries:

lance-csr at SF-300K also carries:

lance-csr at SF-1M also carries:

lance-csr at SF-3M also carries:

SQLite at SF-300K also carries:

SQLite at SF-3M also carries: