Skip to content

Latest commit

 

History

History
107 lines (79 loc) · 2.66 KB

README.md

File metadata and controls

107 lines (79 loc) · 2.66 KB
Raw

gdbb

(Graph DB benchmarks)

Benchmark link-prediction algorithms overs different DBMS

Algorithm information can be found in The Link Prediction Problem for Social Networks

Graphs are extracted from the DBLP Computer Science Bibliography XML, and from Stanford Large Network Dataset Collection

Results can be found here

Prerequisites

Quick Start

  • Fetch code:
git clone https://github.com/natict/gdbb.git
cd gdbb
  • Generate graphs:
./mkgraphs.sh
  • Create MySQL tables and procedures (you'll need mysql root), e.g. for all_core3:
./init-mysql.sh all_core3
  • Execute MySQL benchmark:
./benchmark_mysql.py all_core3
  • Execute Redis benchmark (i.e. for all_core3):
./benchmark_redis.py -d all_core3
  • Create Neo4J graphs (you'll need to be a sudo-er) (i.e. for all_core3):
./init-neo4j.sh all_core3
  • Execute Neo4J benchmark:
./benchmark_neo4j.py all_core3
  • Execute all benchmarks (assuming sudo permissions):
./benchmark-all.sh [MySQL_root_password]

Notes

  • Graph generation is done in-memory, make sure you have enough
  • Some MySQL configuration changes are recommended (significantly speedup load time, allow recursive procedures):
innodb_additional_mem_pool_size = 512M
innodb_buffer_pool_size = 512M
innodb_log_file_size = 256M
innodb_log_buffer_size = 256M
max_sp_recursion_depth=10
  • Using the latest versions of redis-server and redis-py significantly improves performance
  • Currently the Lua scripts are redis-calls optimized and NOT memory optimized
  • Some Redis settings are required:
lua-time-limit 600000	# 10 Minutes
  • Let the Neo4J JVM to allocate more heap (conf/neo4j-wrapper.conf)
# Initial Java Heap Size (in MB)
wrapper.java.initmemory=1024

# Maximum Java Heap Size (in MB)
wrapper.java.maxmemory=16384