Replicating the experiments

• Clone this repo

• Enter to this project root directory

  • cd pmr

• Set PMR_HOME with the path of this project root directory

  • export PMR_HOME=$(pwd)

MillenniumDB

These instructions works on Ubuntu 20.04. Other linux distros might need to install dependencies differently. Libboost version used is 1.71.0, with other versions MillenniumDB might not compile.

• Install MillenniumDB dependencies.

  • sudo apt update
  • sudo apt install g++ cmake libboost-all-dev

• Change directory to the MillenniumDB folder

  • cd MillenniumDB

• Compile MillenniumDB:

  • cmake -Bbuild/Release -DCMAKE_BUILD_TYPE=Release && cmake --build build/Release/

• Create Diamond1000 database:

  • build/Release/bin/create_db ../data/diamond_1000.mdb dbs/diamond_1000

• Create Facebook database

  • build/Release/bin/create_db ../data/facebook.mdb dbs/facebook

• Go back to this project root directory

  • cd ..

• Run diamond1000 benchmark:

  • python3 scripts/benchmark_mdb_diamond.py

• Run facebook benchmark:

  • python3 scripts/benchmark_mdb_facebook.py

Neo4J

• Install Neo4J python driver

  • pip3 install neo4j

• Download and extract the neo4j linux executable: https://neo4j.com/download-center/#community

• Set NEO4J_HOME with the path of the folder extracted

  • export NEO4J_HOME=/path/to/neo4j/folder

• Create facebook database

  $NEO4J_HOME/bin/neo4j-admin import --database facebook \
  --nodes $PMR_HOME/data/facebook_neo4j_nodes.csv \
  --relationships $PMR_HOME/data/facebook_neo4j_edges.csv
  

• Create diamond1000 database

  $NEO4J_HOME/bin/neo4j-admin import --database diamond1000 \
  --nodes $PMR_HOME/data/diamond_1000_neo4j_nodes.csv \
  --relationships $PMR_HOME/data/diamond_1000_neo4j_edges.csv
  

• Edit $NEO4J_HOME/conf/neo4j.conf adding the lines:

  dbms.transaction.timeout=1m
  dbms.default_database=facebook
  dbms.security.auth_enabled=false
  cypher.forbid_shortestpath_common_nodes=false
  

• Run the benchmark with facebook graph:

  • Start the neo4j server

    • bin/neo4j console

  • Wait until the server is ready and run the benchmark in another terminal:

    • python3 scripts/benchmark_neo4j_facebook.py

  • After the benchmark is finished, kill the neo4j server with CTRL-C.

• Run the benchmark with facebook graph:

  • Edit $NEO4J_HOME/conf/neo4j.conf and replace dbms.default_database=facebook with dbms.default_database=diamond1000

  • Start the neo4j server

    • $NEO4J_HOME/bin/neo4j console

  • Wait until the server is ready and run the benchmark in another terminal:

    • python3 scripts/benchmark_neo4j_diamond.py

  • After the benchmark is finished, kill the neo4j server with CTRL-C.

Running WDBench

MillenniumDB

For MillenniumDB we select 576 of the 660 original queries, filtering out those that do not have a fixed node. Queries are at WDBench/sparql_paths_filtered.txt.

• Download the dataset from Figshare
• Uncompress it in the pmr folder:
  • bzip2 -d truthy_direct_properties.nt.bz2
• Transform the .NT file into the MillenniumDB text format:
  • python3 scripts/nt_to_mdb truthy_direct_properties.nt wikidata.mdb
• Create the database in MillenniumDB
  • build/Release/bin/create_db ../data/wikidata.mdb dbs/wikidata
• Execute the benchmarks:
  • python3 scripts/wdbench_paths_mdb.py WDBench/sparql_paths_filtered.txt endpoints
  • python3 scripts/wdbench_paths_mdb.py WDBench/sparql_paths_filtered.txt single_shortest
  • python3 scripts/wdbench_paths_mdb.py WDBench/sparql_paths_filtered.txt all_shortest
  • python3 scripts/wdbench_paths_mdb.py WDBench/sparql_paths_filtered.txt count
  • python3 scripts/wdbench_paths_mdb.py WDBench/sparql_paths_filtered.txt construct_pmr

Neo4J

To load the graph into Neo4J see the original Benchmark repository: https://github.com/MillenniumDB/WDBench#data-loading-for-neo4j

For Neo4J we only considered queries with cypher patterns that can be inside of the operators shortestPath and allShortestPaths. These queries are in the file WDBench/sparql_paths_filtered.txt

To run the benchmark with wikidata graph:

• Edit $NEO4J_HOME/conf/neo4j.conf and set dbms.default_database=wikidata

• Start the neo4j server

  • $NEO4J_HOME/bin/neo4j console

• Wait until the server is ready and run the benchmarks in another terminal:

  • python3 scripts/wdbench_paths_neo4j.py WDBench/sparql_paths_filtered.txt
  • python3 scripts/wdbench_paths_neo4j.py WDBench/sparql_paths_filtered.txt shortestPath
  • python3 scripts/wdbench_paths_neo4j.py WDBench/sparql_paths_filtered.txt allShortestPaths

• After the benchmarks are finished, kill the neo4j server with CTRL-C.