JMH benchmarks to measure the performance of parquet readers
- Apache Parquet Hadoop (parquet-hadoop v1.15.0)
- Apache Arrow Parquet (arrow-dataset v18.1.0)
- Apache Parquet Avro (parquet-avro v1.15.0)
- Make sure you have installed maven and Java 21
mvn packagein the root folder of the project to create JAR a. Copy files from src/main/resources to tmpfs to reduce IO influence, tmpfs is RAM diskjava --add-opens=java.base/java.nio=org.apache.arrow.memory.core,ALL-UNNAMED -jar target/benchmarks.jar -prof gc -p inputPath=#PATH_TO_PARQUET#to run benchmarks
Test data is taken from Taxi & Limousine Commission, TLC Trip Record Data, January 2022, Yellow Taxi Trip Records (https://nyc-tlc.s3.amazonaws.com/trip+data/yellow_tripdata_2022-01.parquet). This file is copied as a resource to the project. Test data contains 2463931 rows and has 19 columns, sample:
+----------+-----------------------+-----------------------+-----------------+---------------+------------+--------------------+--------------+--------------+--------------+-------------+-------+---------+------------+--------------+-----------------------+--------------+----------------------+-------------+
| VendorID | tpep_pickup_datetime | tpep_dropoff_datetime | passenger_count | trip_distance | RatecodeID | store_and_fwd_flag | PULocationID | DOLocationID | payment_type | fare_amount | extra | mta_tax | tip_amount | tolls_amount | improvement_surcharge | total_amount | congestion_surcharge | airport_fee |
+----------+-----------------------+-----------------------+-----------------+---------------+------------+--------------------+--------------+--------------+--------------+-------------+-------+---------+------------+--------------+-----------------------+--------------+----------------------+-------------+
| 1 | 2022-01-01 00:35:40.0 | 2022-01-01 00:53:29.0 | 2.0 | 3.8 | 1.0 | N | 142 | 236 | 1 | 14.5 | 3.0 | 0.5 | 3.65 | 0.0 | 0.3 | 21.95 | 2.5 | 0.0 |
| 1 | 2022-01-01 00:33:43.0 | 2022-01-01 00:42:07.0 | 1.0 | 2.1 | 1.0 | N | 236 | 42 | 1 | 8.0 | 0.5 | 0.5 | 4.0 | 0.0 | 0.3 | 13.3 | 0.0 | 0.0 |
| 2 | 2022-01-01 00:53:21.0 | 2022-01-01 01:02:19.0 | 1.0 | 0.97 | 1.0 | N | 166 | 166 | 1 | 7.5 | 0.5 | 0.5 | 1.76 | 0.0 | 0.3 | 10.56 | 0.0 | 0.0 |
| 2 | 2022-01-01 00:25:21.0 | 2022-01-01 00:35:23.0 | 1.0 | 1.09 | 1.0 | N | 114 | 68 | 2 | 8.0 | 0.5 | 0.5 | 0.0 | 0.0 | 0.3 | 11.8 | 2.5 | 0.0 |
| 2 | 2022-01-01 00:36:48.0 | 2022-01-01 01:14:20.0 | 1.0 | 4.3 | 1.0 | N | 68 | 163 | 1 | 23.5 | 0.5 | 0.5 | 3.0 | 0.0 | 0.3 | 30.3 | 2.5 | 0.0 |
| 1 | 2022-01-01 00:40:15.0 | 2022-01-01 01:09:48.0 | 1.0 | 10.3 | 1.0 | N | 138 | 161 | 1 | 33.0 | 3.0 | 0.5 | 13.0 | 6.55 | 0.3 | 56.35 | 2.5 | 0.0 |
| 2 | 2022-01-01 00:20:50.0 | 2022-01-01 00:34:58.0 | 1.0 | 5.07 | 1.0 | N | 233 | 87 | 1 | 17.0 | 0.5 | 0.5 | 5.2 | 0.0 | 0.3 | 26.0 | 2.5 | 0.0 |
| 2 | 2022-01-01 00:13:04.0 | 2022-01-01 00:22:45.0 | 1.0 | 2.02 | 1.0 | N | 238 | 152 | 2 | 9.0 | 0.5 | 0.5 | 0.0 | 0.0 | 0.3 | 12.8 | 2.5 | 0.0 |
| 2 | 2022-01-01 00:30:02.0 | 2022-01-01 00:44:49.0 | 1.0 | 2.71 | 1.0 | N | 166 | 236 | 1 | 12.0 | 0.5 | 0.5 | 2.25 | 0.0 | 0.3 | 18.05 | 2.5 | 0.0 |
| 2 | 2022-01-01 00:48:52.0 | 2022-01-01 00:53:28.0 | 1.0 | 0.78 | 1.0 | N | 236 | 141 | 2 | 5.0 | 0.5 | 0.5 | 0.0 | 0.0 | 0.3 | 8.8 | 2.5 | 0.0 |
+----------+-----------------------+-----------------------+-----------------+---------------+------------+--------------------+--------------+--------------+--------------+-------------+-------+---------+------------+--------------+-----------------------+--------------+----------------------+-------------+
| Reader | Average, ms | Faster than AvroParquetReader, times | Average gc.alloc.rate, MB/sec | gc.time , ms |
|---|---|---|---|---|
| ArrowParquetReader | 837.026 | 8.457 | 1436.499 | 77 |
| HadoopGroupReader | 3424.390 | 2.067 | 4524.978 | 644 |
| AvroParquetReader | 7078.314 | 1 | 2395.299 | 652 |
Raw results:
Benchmark Mode Cnt Score Error Units
ArrowParquetReaderBenchmark.readAllColumns avgt 5 837.026 ± 8.953 ms/op
ArrowParquetReaderBenchmark.readAllColumns:gc.alloc.rate avgt 5 1436.499 ± 15.323 MB/sec
ArrowParquetReaderBenchmark.readAllColumns:gc.alloc.rate.norm avgt 5 1260805553.333 ± 120820.850 B/op
ArrowParquetReaderBenchmark.readAllColumns:gc.count avgt 5 123.000 counts
ArrowParquetReaderBenchmark.readAllColumns:gc.time avgt 5 77.000 ms
AvroParquetReaderBenchmark.readAllColumns avgt 5 7078.314 ± 202.397 ms/op
AvroParquetReaderBenchmark.readAllColumns:gc.alloc.rate avgt 5 2395.299 ± 67.893 MB/sec
AvroParquetReaderBenchmark.readAllColumns:gc.alloc.rate.norm avgt 5 17777679391.200 ± 562015.622 B/op
AvroParquetReaderBenchmark.readAllColumns:gc.count avgt 5 246.000 counts
AvroParquetReaderBenchmark.readAllColumns:gc.time avgt 5 652.000 ms
HadoopGroupReaderBenchmark.readAllColumns avgt 5 3424.390 ± 92.500 ms/op
HadoopGroupReaderBenchmark.readAllColumns:gc.alloc.rate avgt 5 4524.978 ± 121.529 MB/sec
HadoopGroupReaderBenchmark.readAllColumns:gc.alloc.rate.norm avgt 5 16247558191.467 ± 525162.429 B/op
HadoopGroupReaderBenchmark.readAllColumns:gc.count avgt 5 198.000 counts
HadoopGroupReaderBenchmark.readAllColumns:gc.time avgt 5 644.000 ms
Test data contains 1000 rows and has 45 columns (41 columns for tags and 4 list columns) , sample:
┌─────────────┬───────────────────┬──────────────┬──────────────────┬───┬────────────┬──────────────────────┬──────────────────────┬──────────────────────┬──────────────────────┐
│ application │ availability_zone │ build_number │ cloud_account_id │ … │ tenant_id │ ts │ sums_double │ sums_long │ count │
│ varchar │ varchar │ varchar │ varchar │ │ varchar │ int64[] │ double[] │ int64[] │ int64[] │
├─────────────┼───────────────────┼──────────────┼──────────────────┼───┼────────────┼──────────────────────┼──────────────────────┼──────────────────────┼──────────────────────┤
│ accounts │ zone-b │ 2169 │ 726820024292 │ … │ tenant-267 │ [1737967143010, 17… │ [] │ [26, 36, 45, 93, 1… │ [8, 16, 21, 25, 34… │
│ accounts │ zone-b │ 2604 │ 299380463457 │ … │ tenant-488 │ [1737967143010, 17… │ [] │ [5, 10, 36, 73, 11… │ [3, 13, 17, 19, 23… │
│ inventory │ zone-a │ 9821 │ 101872852789 │ … │ tenant-953 │ [1737967143010, 17… │ [] │ [38, 80, 110, 139,… │ [7, 17, 23, 31, 35… │
│ accounts │ zone-c │ 2876 │ 748866464070 │ … │ tenant-344 │ [1737967143010, 17… │ [] │ [35, 78, 83, 125, … │ [6, 13, 19, 28, 34… │
| Reader | Average, ms | Faster than AvroParquetReader, times | Average gc.alloc.rate, MB/sec | gc.time , ms |
|---|---|---|---|---|
| ArrowParquetReader | 4.047 | 9.650 | 2111.226 | 65 |
| HadoopGroupReader | 23.331 | 1.674 | 4238.102 | 436 |
| OptimizedReader | 16.819 | 2.322 | 2542.464 | 356 |
| AvroParquetReader | 39.054 | 1 | 1697.357 | 229 |
Raw results:
Benchmark (inputPath) Mode Cnt Score Error Units
ArrowParquetReaderBenchmark.readAllColumns /tmp/syntetic.parquet avgt 5 4.047 ± 0.619 ms/op
ArrowParquetReaderBenchmark.readAllColumns:gc.alloc.rate /tmp/syntetic.parquet avgt 5 2111.226 ± 319.393 MB/sec
ArrowParquetReaderBenchmark.readAllColumns:gc.alloc.rate.norm /tmp/syntetic.parquet avgt 5 8947556.587 ± 3819.001 B/op
ArrowParquetReaderBenchmark.readAllColumns:gc.count /tmp/syntetic.parquet avgt 5 190.000 counts
ArrowParquetReaderBenchmark.readAllColumns:gc.time /tmp/syntetic.parquet avgt 5 65.000 ms
AvroParquetReaderBenchmark.readAllColumns /tmp/syntetic.parquet avgt 5 39.054 ± 2.273 ms/op
AvroParquetReaderBenchmark.readAllColumns:gc.alloc.rate /tmp/syntetic.parquet avgt 5 1697.357 ± 97.093 MB/sec
AvroParquetReaderBenchmark.readAllColumns:gc.alloc.rate.norm /tmp/syntetic.parquet avgt 5 69497445.903 ± 40605.182 B/op
AvroParquetReaderBenchmark.readAllColumns:gc.count /tmp/syntetic.parquet avgt 5 281.000 counts
AvroParquetReaderBenchmark.readAllColumns:gc.time /tmp/syntetic.parquet avgt 5 229.000 ms
HadoopGroupReaderBenchmark.readAllColumns /tmp/syntetic.parquet avgt 5 23.331 ± 2.642 ms/op
HadoopGroupReaderBenchmark.readAllColumns:gc.alloc.rate /tmp/syntetic.parquet avgt 5 4238.102 ± 461.358 MB/sec
HadoopGroupReaderBenchmark.readAllColumns:gc.alloc.rate.norm /tmp/syntetic.parquet avgt 5 103616114.393 ± 36254.937 B/op
HadoopGroupReaderBenchmark.readAllColumns:gc.count /tmp/syntetic.parquet avgt 5 472.000 counts
HadoopGroupReaderBenchmark.readAllColumns:gc.time /tmp/syntetic.parquet avgt 5 436.000 ms
optimized.OptimizedReaderBenchmark.readAllColumns /tmp/syntetic.parquet avgt 5 16.819 ± 1.211 ms/op
optimized.OptimizedReaderBenchmark.readAllColumns:gc.alloc.rate /tmp/syntetic.parquet avgt 5 2542.464 ± 179.210 MB/sec
optimized.OptimizedReaderBenchmark.readAllColumns:gc.alloc.rate.norm /tmp/syntetic.parquet avgt 5 44827908.387 ± 24944.764 B/op
optimized.OptimizedReaderBenchmark.readAllColumns:gc.count /tmp/syntetic.parquet avgt 5 413.000 counts
optimized.OptimizedReaderBenchmark.readAllColumns:gc.time /tmp/syntetic.parquet avgt 5 356.000 ms