Closed sampollard closed 7 years ago
Looking through the source code of the Graphalytics/PowerGraph implementation, HDFS is not used or supported as data source when running Graphalytics. Instead, the graph data is read directly from the directory given by graphs.root-directory
on the machine running the Graphalytics benchmark process. PowerGraph then distributes the data internally.
To fix the compilation issues, I think PowerGraph needs to be compiled with --no-jvm
.
@stijnh As main author of this implementation, can you confirm the above? If so, I will update the README to reflect this.
Got it. I compiled PowerGraph with the following commands:
git checkout https://github.com/sampollard/PowerGraph
cd PowerGraph
./configure --no_jvm
cd /release/toolkits
make
After this works,
cd graphalytics-platforms-powergraph
mvn package
tar -xf graphalytics-0.3-powergraph-0.1
cd graphalytics-0.3-powergraph-0.1
# Assuming you have already configured graphalytics
cp -r /path/to/ldbc_graphalytics/config/ config
cp config-template/powergraph.properties config/
# Configure powergraph.properties as per the template
./run-benchmark
I think you also need to have the benchmark.properties in the same directory as powergraph.properties.
After packaging with
mvn package -DskipTests
, running./run-benchmark.sh
yieldsThis can be (temporarily) patched by editing
bin/standard/CMakeFile/main.dir/link.txt
to include the-lhdfs
flag. However, running the benchmark again yieldsNotice that this is with my LD_LIBRARY_PATH set to /usr/lib/jvm/java-8-openjdk-amd64/jre/lib/amd64/server/