Re: [igraph] Better layout output in igraph

[Tamas Nepusz]

> Now that you point out this fact... is there any comparison between the 
> performance of Gephi vs R laying out graphs?
I don't think there is, partly because Gephi has a GUI so it has some
overhead associated with drawing the graph continuously while it is being
laid out. (I'm not _that_ familiar with Gephi, so things may have changed
now, but I remember that when I tested it, it was re-drawing the graph from
time to time while the layout algorithm was being run). However, one thing
that I know is that the DrL layout meant for _really_ large graphs in igraph
is practically the same as the OpenOrd layout in Gephi -- formerly the
algorithm was called DrL when we lifted it into igraph. So there should be
no big performance difference between the two, and I expect that the
differences can mostly be accounted for

1) the continuous redrawing in Gephi and
2) the fact that it's written in Java

The latter means that someone had to port the original C++ code of DrL into
Java and that might have affected the performance, probably either way -- I
have seen quite a few articles claiming that Java is more efficient than  C
or C++ after the JIT compiler managed to optimize the hotspots, but I still
find it somewhat hard to believe and no one tested it on network science
problems as far as I know.

If anyone is willing to conduct a benchmark between Gephi and igraph, I'm
really interested in the results -- at least we would know which parts of
igraph to concentrate the optimization efforts on. :)

-- 
T.