Does it has better settings for presto-native-execution? Test the C++ Worker TPC-DS performance is not better compare with Java Woker

[yixi-gu]

Environment Description

• Presto version used: 0.287-SNAPSHOT
• Storage (HDFS/S3/GCS..): HDFS （remote hdfs storage cluster）
• Data source and connectors or catalogs used: hive-hadoop2
• Deployment: Could ECS, 12 * 2xlarge (8c64g)

C++ Woker PrestoServer settings:

config.properties:

http-server.http.port=8889 experimental.spiller-spill-path=/data/data1/presto presto.version=0.287 discovery.uri=http://master:8889 exchange.http-client.request-timeout=10s query.max-memory-per-node=40GB coordinator=false memory.heap-headroom-per-node=2GB

velox.properties: mutable-config=true

Java Woker PrestoServer settings: config.properties:

http-server.http.port=8889 node-manager.http-client.max-content-length=64MB coordinator=false query.max-stage-count=200 discovery.uri=http://master:8889 exchange.http-client.max-content-length=128MB exchange.http-client.request-timeout=10s jmx.rmiserver.port=19081 query.max-memory-per-node=20GB log.max-size=1000MB experimental.max-revocable-memory-per-node=40GB query.max-total-memory-per-node=20GB experimental.spiller-spill-path=/data/data1/presto experimental.reserved-pool-enabled=false jmx.rmiregistry.port=19080 memory.heap-headroom-per-node=10GB experimental.spill-enabled=true

[andugog]

What is the size of the dataset?