wrong number of iterations

[alex031029]

I tested Cusha on LiveJournal dataset, finding only 1 iteration performed. PR, BFS and SSSP all had this issue. Is that a problem of wrong input format?I used edgelist as the input. Thanks!

[alex031029]

Besides, some graphs will be divided into 0 shards. I am not sure what's wrong with my set up.

[farkhor]

The simplest execution of CuSha in Ubuntu 14.04 with $ ./cusha --input soc-LiveJournal1.txt --method CW and with BFS specified as the algorithm results in 10 iterations while the input graph is divided into 789 shards. Can you please be more specific about your configuration and how you run the framework? I also suggest you to read the usage as well as existing discussed (and solved) issues carefully; then start from the simplest algorithm (BFS).

[alex031029]

Thanks for your response. I ran Cusha under Ubuntu 14.04 with CUDA 7.0. I also got 789 shards, while only 1 iteration was performed. I also ran belgium_osm and road_usa, which turned out to be 0 shards.

This is how I ran Cusha

1. Uncommented the application in user_specified_global_configuration.h
2. Preprocess the graph for PageRank using the program you offered in first issue #1 .
3. $ sh sfBuild.sh
4. $ ./cusha --input soc-LiveJournal1.txt --method CW

I downloaded the the datasets from http://www.cise.ufl.edu/research/sparse/MM/DIMACS10/belgium_osm.tar.gz http://www.cise.ufl.edu/research/sparse/MM/SNAP/soc-LiveJournal1.tar.gz http://www.cise.ufl.edu/research/sparse/MM/DIMACS10/road_usa.tar.gz

[farkhor]

Please start with the simplest algorithm (BFS) by un-commenting the corresponding line in user_specified_global_configurations.h. It seems to me that you want to run PageRank but you're using the same original edge-list, not the generated file (in your second step).

[alex031029]

Thanks for your kind help. And sorry that I didn't express well. The second step is optional, I only preprocess for PageRank application, and of course use the generated file as the input for Cusha.

And I have tried all SSSP, PageRank and BFS. Unfortunately, none of them return expected answer. BFS application also gives me 1 iteration and 789 shards for soc-LiveJournal, and 0 shard for Belgium_osm and road_usa.

[farkhor]

Have you tried running with a different arbparam input as the source node for BFS? Maybe the structure of the graphs you're using have different indices? Can you also profile if the CUDA kernel gets executed for you at all?

[alex031029]

That's it! After I changed arbparam to 2 or 3, it returned correct result. For Belgium_osm and road_usa, I just found I forgot to set the paramater to undirected. Many thanks for your help!