TeraHeap: Reducing Memory Pressure in Managed Big Data Frameworks

Description

TeraHeap extends the managed runtime (JVM) to use a second, high-capacity heap over a fast storage device that coexists with the regular heap. TeraHeap provides direct access to objects on the second heap (no S/D). It also reduces GC cost by fencing the garbage collector from scanning the second heap. TeraHeap leverages frameworks’ property of choosing specific objects for off-heap placement and offers frameworks a hint-based interface for moving such objects to the second heap.

Install Prerequisites

Install the following packages:

sudo yum install python3-pip
pip3 install scan-build --user
pip3 install compdb --user

OpeJDK1.8 can be compiled with OpenJDK1.8. So, you need first to download OpenJDK1.8.

# For Ubuntu
sudo apt-get install openjdk-8-jdk

# For Centos
sudo yum install java-1.8.0-openjdk

Edit the compile.sh script in jdk8u345/ directory by changing the JAVA_HOME variable to point to the installation directory of openjdk8 as follows:

export JAVA_HOME="/path/to/openjdk8"

The JVM can be compiled only with gcc <= 8.5 and g++ <= 8.5

Build

1. Build allocator.

cd allocator
./build.sh
cd -

Read the README.md file in allocator directory to export the specific environment variables

2. Build tera_malloc.

cd tera_malloc
./build.sh
cd -

Read the README.md file in tera_malloc directory to export the specific environment variables

3. Set your gcc/g++ path/alias

cd ./jdk8u345 # for building java8
cd ./jdk17u067 # for building java17

and set CC and CXX variables inside compile.sh to your gcc path/alias

4. Build JVM (release mode)

./compile.sh -r
cd -

or

Build JVM (fastdebug mode)

./compile.sh -d
cd -

Benchmarks

To run benchmarks please clone the repository tera_applications and read the README.md files in each application directory. There are instructions about how to compile and run the different applications.

Acknowledgements

We thank the anonymous reviewers for their insightful comments and their help in preparing the final version of the paper. We thankfully acknowledge the support of the European Commission under the Horizon 2020 Framework Programme for Research and Innovation through the projects EVOLVE (grant agreement No 825061). This research is also partly supported by project EUPEX, which has received funding from the European High-Performance Computing Joint Undertaking (JU) under grant agreement No 101033975. The JU receives support from the European Union's Horizon 2020 re-search and innovation programme and France, Germany, Italy, Greece, United Kingdom, Czech Republic, Croatia. Iacovos G. Kolokasis is also supported by the Meta Research PhD Fellowship and the State Scholarship Foundation of Cyprus.