XPGraph inserts edges faster than DGAP when using 1 writer thread with Orkut

[ydsungan]

Run XPGraph with the command _./bin/main -f ${DATA_PATH}/xpgraph/orkut/ -p0 ${PMEM_PATH}/xpgraph-db/XPGraph0/ --recovery ${PMEMPATH}/xpgraph-db/XPGraphRecovery/ --source 1 -v 3072627 -q 5 -j 0 -t 1 (-j: 0 for ingest, 1 for log, 2 for archive, 3 for recover). It can insert the Orkut in 10 seconds.

While running DGAP with the command export OMP_NUM_THREADS=1 ./bfs -B ${DATA_PATH}/orkut/base.el -D ${DATA_PATH}/orkut/dynamic.el -f ${PMEM_PATH}/orkut.pmem -r 1 -n 5 -a, it takes 127 seconds to insert the Orkut.

The comparison results are completely different from those in the paper.

[biqar]

Thank you for your feedback, and I'm sorry to hear that you faced a performance issue in your test. In our experiment, we observed that DGAP required approximately 83 seconds, while XpGraph required around 180 seconds to insert the Orkut graph. Could you please specify:

1. How you prepared your datasets?
2. Which job ID you used to run the XpGraph test? Please ensure you used -j 2 to run tests in XpGraph.

[ydsungan]

There are 117185083 lines in the dataset Orkut. The split_dataset.cc generates two directed edges for each line. edge_list[line++] = make_pair(u, v); edge_list[line++] = make_pair(v, u);

Then use the base and dynamic graph files generated by split_dataset.cc as the input of the preprocessor text2bin.

However, XPGraph itself will also insert a dataset as an undirected graph. As shown in the picture below, XPGraph has 234370166 in-edges and 234370166 out-edges and takes 178 seconds to insert the Orkut.

If I use the original Orkut dataset (el format) as the input of the preprocessor text2bin, XPGraph can insert the Orkut correctly. As shown in the picture below, XPGraph has 117185083 in-edges and 117185083 out-edges and takes 27 seconds to insert the Orkut.

I used -j 2 to run tests in XPGraph

[biqar]

Thanks for your detailed feedback, @ydsungan. Unfortunately, I am traveling and won't be able to verify the issue now. I will get back to you soon.