Re: [igraph] igraph-help Digest, Vol 93, Issue 6

[Gábor Csárdi]

On Mon, Apr 7, 2014 at 12:21 PM, Bian, Jiang <address@hidden> wrote:

[...]

For my purpose, obviously, it’s the calculations that matter to me. So, handle means handle the calculations. Sorry for being not clear.

"Calculations" is not any clearer, unfortunately. It depends on exactly what kind of calculations. igraph has hundreds of functions, some of these only work with networks up to a couple of hundred vertices. So you would say it can only handle graphs with 100 vertices? You are free to do so, but I don't think this is fair.

> And you'll never find any piece of software that can.

Some of the new graph-parallel processing systems (such as Graphlab’s PowerGraph, Apache’s Giraph, etc.) can, but they are just not as complete as igraph. For example, Graphlab only has PageRank and Triangle Counting implemented.

No, they can not. igraph can calculate PageRank and triangle counting easily for your graphs, most probably faster than GraphLab. You are saying that these systems can handle your graphs, because they only implement a handful of the "easy" algorithms. This does not make sense to me.

So, I guess I am on my own to implement all the calculations in one of the HPC frameworks. I was trying to avoided it, since it wasn’t really the goal of my study.

Indeed. Btw. I would not expect that betweenness in GraphLab will be as fast as triangle counting. It won't. I guess for your graphs it will be about 50,000 times slower. Which might still be good enough.

Gabor

[...]