Server benchmarking results for the paper
General Info
- Machine: AWS EC2 c5.2xlarge instance
- CPU: Intel Xeon Platinum 8275CL (8) @ 3.599GHz
- RAM: 16 GB
- OS: Amazon Linux 2 x86_64
- Kernel: 5.10.144-127.601.amzn2.x86_64
- python: Python 3.10.5 | packaged by conda-forge | (main, Jun 14 2022, 07:04:59) [GCC 10.3.0] on linux
Results were generated by downloading all the datasets and running ./paper.sh.
The file bench-results-server.db includes info of the graph datasets this benchmark uses and the benchmarking results.
In the bench_results table in the sqlite database, records with id from 1-96 were generated by /mentrypoint_paper.sh called by ./paper.sh,
while records with id from 97-300 were generated by ./entrypoint_paper.sh called by ./paper.sh.
Benchmarking easygraph multiprocessing
The /mentrypoint_paper.sh file runs scripts that measures performance of easygraph multiprocessing methods
Methods benchmarked:
# in config.py
easygraph_multipcoessing_methods_for_paper = {
'betweenness_centrality',
'closeness_centrality',
'constraint',
'hierarchy',
}On these datasets:
# real-world network
dataset_names_for_paper_multiprocessing = [
'bio',
'uspowergrid',
'enron',
'coauthorship',
]
# and erdos-renyi random networks (undirected)
er_dataset_names_for_paper_multiprocessing = [
f'er_{x}' for x in (500, 1000, 5000, 10000)
]
# additions from the 20221213 runs:
er_dataset_names_for_paper_20221213 = [
f'er_{x}'
for x in (50000, 500000, 1000000)
]Benchmarking easygraph with and without C++ binding, and networkx
Done by ./entrypoint_paper.sh.
average_time = -1.0 means that the method is not supported by that graph type (e.g., undirected graph)
Methods benchmarked:
clustering_methods = ["clustering"]
shortest_path_methods = [('Dijkstra', 'single_source_dijkstra_path')] # eg.Dijkstra() vs nx.single_source_dijkstra_path()
connected_components_methods_G = [
"connected_components",
"biconnected_components",
'strongly_connected_components',
]
mst_methods = ['minimum_spanning_tree']
other_methods = ['density', 'constraint']
new_methods = ['effective_size']With these multiprocessing options:
# n_workers not specified
# and 2, 4
easygraph_multiprocessing_n_workers_options_for_paper = [2, 4]On these datasets:
[
"cheminformatics", # read world networks
"bio",
"eco",
'pgp',
'pgp_undirected',
'road',
'uspowergrid',
'enron',
'coauthorship',
'stub',
'stub_with_underscore',
'stub_directed',
'stub_nx',
] + [f'er_{x}' for x in (500, 1000, 5000, 10000)] # erdos-renyi random networks (undirected)Bench results REST API
- Try it out:
- For local benchmarking results:
- For server benchmarking results:
Just prepend/sin the URL path, like so:
- Documentation: https://easygraph-bench-results.teddysc.me/docs