ClickBench — analytical databases

Analytical-database runtimes for 43 SQL queries over a ~100M-row web-analytics dataset.

The latest published result per system and machine from ClickBench (github.com/ClickHouse/ClickBench), a benchmark of analytical databases over 43 fixed queries on a ~100M-row web-analytics dataset. Load time and data size are as recorded in the source's result files; the cold and hot totals are sums, computed here, of the source's 43 per-query timings (cold = each query's first run, hot = the best of its second and third). Lower is faster. The source's own headline 'relative' score is not reproduced.

Results

hot_total_s (s) per subject over 778 measurements, ranked by average.

SubjectAvgMedianMinMaxP95P99Count
infobright-c6a-4xlarge469,132469,132469,132469,132469,132469,1321
mariadb-c6a-4xlarge147,906147,906147,906147,906147,906147,9061
yugabyte-c6a-4xlarge51,70251,70251,70251,70251,70251,7021
Aurora for MySQL (Aurora: 16acu)28,86028,86028,86028,86028,86028,8601
mysql-c6a-4xlarge21,96221,96221,96221,96221,96221,9621
postgresql-orioledb-c6a-4xlarge20,06520,06520,06520,06520,06520,0651
sqlite-c6a-large19,92019,92019,92019,92019,92019,9201
mongodb-c6a-4xlarge19,43019,43019,43019,43019,43019,4301
mysql-myisam-c6a-2xlarge19,23419,23419,23419,23419,23419,2341
mysql-myisam-c6a-4xlarge18,32018,32018,32018,32018,32018,3201
pg-duckdb-c6a-xlarge14,97614,97614,97614,97614,97614,9761
sqlite-c6a-4xlarge12,85912,85912,85912,85912,85912,8591
sqlite-c6a-xlarge12,43312,43312,43312,43312,43312,4331
sqlite-c6a-2xlarge12,41712,41712,41712,41712,41712,4171
postgresql-c6a-4xlarge11,89611,89611,89611,89611,89611,8961
pg-duckdb-c8g-4xlarge11,53011,53011,53011,53011,53011,5301
pg-duckdb-c6a-4xlarge11,52511,52511,52511,52511,52511,5251
pg-duckdb-c6a-2xlarge11,30411,30411,30411,30411,30411,3041
cockroachdb-c6a-4xlarge10,96610,96610,96610,96610,96610,9661
cockroachdb-c8g-4xlarge10,39010,39010,39010,39010,39010,3901
hive-parquet-single-c6a-4xlarge9,3859,3859,3859,3859,3859,3851
timescaledb-no-columnstore-c8g-4xlarge8,8548,8548,8548,8548,8548,8541
timescaledb-no-columnstore-c6a-4xlarge8,8278,8278,8278,8278,8278,8271
sail-parquet-t3a-small7,1397,1397,1397,1397,1397,1391
sail-parquet-partitioned-t3a-small6,9106,9106,9106,9106,9106,9101

…and 750 more subjects.

Metrics

  • hot_total_s (s) — recorded on each measurement. Total hot runtime: for each of the 43 benchmark queries, the best of the second and third runs, summed, in seconds. Lower is better — this is the headline number. Queries the system could not run are left out of the sum
  • cold_total_s (s) — recorded on each measurement. Total cold runtime: each query's first run (caches unprimed), summed, in seconds. Lower is better. Failed queries are left out of the sum and counted as missing_queries in the measurement metadata
  • load_time_s (s) — recorded on each measurement. Time to load the benchmark dataset into the system, in seconds. Zero for systems that query the source files in place without a load step
  • data_size_bytes (bytes) — recorded on each measurement. Size of the dataset once loaded into the system, in bytes — smaller means better compression
  • concurrent_qps (queries/s) — recorded on each measurement. Sustained queries per second in ClickBench's concurrency test, where the system was measured. Higher is better

Subjects (778)

  • AlloyDB (128GiB)
  • AlloyDB (64GiB)
  • AlloyDB (tuned) (64GiB)
  • Arc (c6a.2xlarge)
  • Arc (c6a.4xlarge)
  • Arc (c6a.large)
  • Arc (c6a.metal)
  • Arc (c6a.xlarge)
  • Arc (c7a.metal-48xl)
  • Arc (c8g.4xlarge)
  • Arc (c8g.metal-48xl)
  • Arc (t3a.small)
  • Athena (partitioned) (serverless)
  • Aurora for MySQL (Aurora: 16acu)
  • Aurora for PostgreSQL (Aurora: 16acu)
  • Bigquery (serverless)
  • BQN (c6a.4xlarge)
  • BQN (c6a.metal)
  • BQN (c7a.metal-48xl)
  • BQN (c8g.4xlarge)
  • BQN (c8g.metal-48xl)
  • ByConity (c6a.4xlarge)
  • CedarDB (c6a.2xlarge)
  • CedarDB (c6a.4xlarge)
  • CedarDB (c6a.metal)
  • CedarDB (c6a.xlarge)
  • CedarDB (c7a.metal-48xl)
  • CedarDB (c8g.4xlarge)
  • CedarDB (c8g.metal-48xl)
  • CedarDB (Parquet) (c6a.2xlarge)
  • CedarDB (Parquet) (c6a.4xlarge)
  • CedarDB (Parquet) (c6a.large)
  • CedarDB (Parquet) (c6a.metal)
  • CedarDB (Parquet) (c6a.xlarge)
  • CedarDB (Parquet) (c7a.metal-48xl)
  • CedarDB (Parquet) (c8g.4xlarge)
  • CedarDB (Parquet) (c8g.metal-48xl)
  • CedarDB (Parquet) (t3a.small)
  • chDB (c6a.2xlarge)
  • chDB (c6a.4xlarge)
  • chDB (c6a.large)
  • chDB (c6a.metal)
  • chDB (c6a.xlarge)
  • chDB (c7a.metal-48xl)
  • chDB (c8g.4xlarge)
  • chDB (c8g.metal-48xl)
  • chDB (DataFrame) (c6a.metal)
  • chDB (DataFrame) (c7a.metal-48xl)
  • chDB (DataFrame) (c8g.metal-48xl)
  • chDB (Parquet, partitioned) (c6a.2xlarge)

…and 728 more.

Published by ClickBench.