This tests random and sequence HDFS I/O.

In order to build and run this test, you must put the HDFS and Hadoop-common jar files into your classpath. In the Hadoop install, these are found under share/hadoop/common/ share/hadoop/hdfs/

Alternatively, you can generate a jar using maven:

mvn package

TODO: use Ivy.

You have to set some Java system properties when running the test.

Here is an example of how to run the test:

ANT\_OPTS="-Dhio.nthreads=5 -Dhio.ngigs.to.read=10 -Dhio.ngigs.in.file=10 -Dhio.hdfs.uri=hdfs://localhost:4000" ant compile jar run

It's also important to make sure libhadoop.so is in your LD_LIBRARY_PATH; otherwise, you won't get features like short-circuit local reads which are important for performance.

Alternately, you can directly run the jar with:

java -Dhio.nthreads=5 -Dhio.ngigs.to.read=1 -Dhio.ngigs.in.file=1 -Dhio.hdfs.uri=hdfs://localhost:6000 com.cloudera.HioBench

You can also run the jar using 'hadoop jar' command:

First, set the required configurations

export HADOOP_OPTS="-Dhio.nthreads=1 -Dhio.ngigs.to.read=4 -Dhio.read.chunk.bytes=1048576 -Dhio.hdfs.uri=hdfs://localhost:9000/ -Dhio.hdfs.file.name=hdfs://localhost:9000/user/host/test.in -Dhio.ngigs.in.file=4 -Dhio.hdfs.test.type=random -Dverbose"

Then run the benchmark

hadoop jar hio_test-0.0.1-SNAPSHOT.jar com.cloudera.HioBench

Colin Patrick McCabe [email protected]

Ahmed Mahran [email protected]