4139 - Presto C++ 的学习与基准测试.pdf

1、Hands-On Workshop:Diving into Presto C+Pramod SatyaTalk OutlinePre-reqsPresto C+/Velox overviewPbench overviewHands on labLatest performance benchmarksFuture roadmaphttps:/ibm.github.io/presto-native-lab/Pull these images beforehanddocker pull prestodb/presto:latestFor Mac silicon(M1,etc)docker pull

2、 public.ecr.aws/oss-presto/presto-native:0.289-ubuntu-arm64For Linuxdocker pull public.ecr.aws/oss-presto/presto-native:0.288Download the TPC-DS/TPC-H data from here:https:/presto-virtual- Java Cluster ArchitectureCoordinatorParserOptimizerSchedulerHive MetastoreWorkerExecutionHive ConnectorWorkerEx

3、ecutionHive ConnectorWorkerExecutionHive ConnectorWorkerExecutionHive ConnectorS3(ORC,Parquet)Tasks,Monitoring,Serialized PageSerialized PageSerialized PageSerialized PageJava coordinator and workersGC pausesPerformance cliffsOperational difficultiesHard to control bare metal level performanceWhat i

4、s Presto C+?Presto C+is a full re-write of the Presto worker in C+Implements Prestos HTTP interface,including:Co-ordinator to worker orchestrationExecution of Presto plan fragments+splitsWorker-to-worker exchange protocolStatus reporting end-points Drop in replacement for Presto Java worker Leverage

5、s Velox ref:Velox Metas Unified Execution Engine(VLDB 2022)Presto C+Cluster ArchitectureC+workerBuilt over the Velox librarySIMDRuntime optimizationsSmart I/O prefetching/cachingMemory Arbitration featuresCoordinatorParserOptimizerSchedulerHive MetastoreS3(ORC,Parquet)Tasks,Monitoring,Serialized Pag

6、eSerialized PageSerialized PageSerialized PagePresto C+VeloxVelox Hive ConnectorPresto C+VeloxVelox Hive ConnectorPresto C+VeloxVelox Hive ConnectorPresto C+VeloxVelox Hive ConnectorBrief Overview of VeloxVelox is an open source unified execution engine that helps to consolidate data management syst