Relatively large number of edges to nodes

[griffex]

The files I'm working with contain around 1,000,000 directed edges and about 11,000 nodes. I know openord is supposed to work well for large graphing project, but I'm concerned its edge cutting will skew my clusters, particularly since the graph is directed. Does anyone have alternative recommendations on layouts that might be well balanced for this type of graph?

[francis_flavin]

Hi griffex,

That's quite a graph you have--very impressive!

My favorite algorithm that balances speed and layout aesthetics is Force Atlas 2. It lays out the graph in ways that work very well for many of the graphs I analyze and it usually works well with dynamic graphs. However, I've never tried it on a graph with 1,000,000 edges.

Just out of curiosity, what type of hardware (CPU, GPU, RAM, etc.) are you using for million-edge graphs? And, are you running Gephi on a 64-bit Java platform or a 32-bit Java platform? I'm always eager to hear hardware evaluations from people who are analyzing big graphs in Gephi.

Thanks--and good luck with your project!


Francis

[griffex]

Thanks, I'm attempting to map an internal linking structure of a large website to analyze pagerank flow and clustering. It's a bit of a bear. I'm definitely running a 64-bit Java platform. PC specs are i7 930 oc'd to 3.2, 6GB ram Kingston 2100hz (assigned 3500MB to heap I think), dual 470GTX in SLI. Seemed to process everything out alright. Force atlats 2 didn't work out too well, compacted everything tightly. Ended up using standard force atlas on about a 10,000 repulsion with something like 200 attraction force. Left it running for about 5-7 hours and it turned out alright.