What

A rig for comparing the performance of various hash tables in Java. Used to produce the results for this blog post on hash table 'services' and this blog post on Java hash table 'libraries'.

It's neither the prettiest code, nor a rigorous benchmark. It was written to get a solid feel for the relative performance of these packages for our particular workload. See the links above for more details.

Setup

• Go to EC2 and provision a m2.4xlarge instance with AMI ami-31d41658 (Standard Redhat 6.1 64-bit).
• ssh in as root.
• Download Maven 3.0.x binaries from a mirror and install per the instructions.

If you want to set up the hash table 'services' comparison:

• yum install gcc-c++ zlib-devel java-1.6.0-openjdk-devel.x86_64

• export JAVA_HOME=/usr/lib/jvm/java-1.6.0-openjdk.x86_64

• Download the sources for the data stores and install them:

  • redis-2.4.2

        make
        make install
    

  • kyotocabinet-1.2.70

        ./configure
        make
        make install
    

  • kyotocabinet-java-1.21

        # make sure you have $JAVA_HOME set appropriately!
        ./configure
        make
        make install
    

  • berkeley-db-5.2.36 (You'll need to log in to get this one.)

        cd build_unix
        ../dist/configure --enable-java
        make
        make install
    

• Add /usr/local/lib/ to /etc/ld.so.conf so that Java can see the shared Kyoto Cabinet libraries.

• Run a ldconfig to make sure the links and caches are fresh.

If you want to set up the hash table 'libraries' comparison:

• yum install java-1.6.0-openjdk-devel.x86_64

Run

• Get the test code source

    git clone git://github.com/aggregateknowledge/hashperf.git
  

• Grab dependencies and compile

    cd hashperf
    mvn compile
    mkdir deps
    mvn dependency:copy-dependencies -DoutputDirectory=deps
  

If you want to run the hash table 'services' comparison:

• Start the daemons

    /path/to/redis-server /path/to/hashperf/redis.conf
  

• Run!

    java -server -Djava.library.path=/usr/local/lib/:/usr/local/BerkeleyDB.5.2/lib/ -classpath deps/*:target/classes net.agkn.hashperf.services.FullPerformanceTestSuite /path/to/data.csv /path/to/stats/dir/ warmupCount obsCount pollingInterval
  

  in my case, this was:

    mkdir /dev/shm/stats/
    mv data.csv /dev/shm/hash_test.csv
    java -server -Djava.library.path=/usr/local/lib/:/usr/local/BerkeleyDB.5.2/lib/ -classpath deps/*:target/classes net.agkn.hashperf.services.FullPerformanceTestSuite /dev/shm/hash_test.csv /dev/shm/stats/ 10 30 1000000
  

Note that the line count in your test file divided by the pollingInterval should be less than or equal to net.agkn.hashperf.util.RecordIterator#POLL_COUNT.

If you want to run the hash table 'libraries' comparison:

• Run!

    java -server -XmxNNNg -classpath deps/*:target/classes net.agkn.hashperf.libs.PerformanceTestSuite /path/to/data.csv /path/to/stats/dir/ warmupCount obsCount pollingInterval sizeHint
  

  in my case, an example of this was:

    mkdir /dev/shm/stats/
    mv data.csv /dev/shm/hash_test.csv
    java -server -Xmx50g -classpath deps/*:target/classes net.agkn.hashperf.libs.PerformanceTestSuite /dev/shm/hash_test.csv /dev/shm/stats/ 2 2 10000000 976000000
  

License

This program and the accompanying materials are made available under the terms of the Eclipse Public License v1.0 which accompanies this distribution, and is available at http://www.eclipse.org/legal/epl-v10.html.

Contributors:

Aggregate Knowledge - implementation