Hadoop HUJI: Measurement Hook-Points

From Lawa
Jump to: navigation, search

Hadoop 0.21 Hook Points

1. Mapper Input:

org.apache.hadoop.hdfs.BlockReader.readChunk()

2. Reducer Input:

org.apache.hadoop.mapreduce.task.reduce.Fetcher.shuffleToMemory()
org.apache.hadoop.mapreduce.task.reduce.Fetcher.shuffleToDisk()

3. HDFS Writes:

org.apache.hadoop.hdfs.server.datanode.BlockReceiver.receiveBlock()

Hadoop 0.20.2 (CDH3) Hook Points

1. Mapper Input

hdfs.org.apache.hadoop.hdfs.DFSClient.BlockReader.readChunk()

Instrumentation of the input stream is in the factory method public static BlockReader newBlockReader (Same as Hadoop 0.21 )

2. Reducer Input:

mapred.org.apache.hadoop.mapred.ReduceTask.ReduceCopier.MapOutputCopier.shuffleInMemory()
mapred.org.apache.hadoop.mapred.ReduceTask.ReduceCopier.MapOutputCopier.shuffleToDisk()

3. HDFS Writes:

hdfs.org.apache.hadoop.hdfs.server.datanode.BlockReceiver.receiveBlock()

The following diagram helps illustrate the locations of these hook-points in the execution flow of the job: Hadoop-statistics-hookpoints.jpg