igraph-help
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [igraph] Is it possible to use igraph in Java?


From: Jeff G
Subject: Re: [igraph] Is it possible to use igraph in Java?
Date: Fri, 3 Apr 2009 09:09:08 -0400

I really like NetworkX, they have a good package and they're a responsive group.  Their performance on spring graphs (FR) has been significantly improved recently by using a numpy matrix/sparse matrix.  What use to take a week to plot, now takes an hour (20,000 nodes).  However, FR is not laying out well for me on large graphs, so I'm very interested in DrL and LGL.  NetworkX does indeed have a component for connecting to UbiGraph.  That is where I heard about the scalability, where someone stated "The most I have displaying well in Ubigraph is around 2000 nodes and 2500 edges."  I'm excited to test out the performance of igraph and see what the layouts produce.  I will need 3d though, so I can't wait until you guys and make that happen (fingers crossed).  :-)

- Jeff

On Fri, Apr 3, 2009 at 3:18 AM, Tamas Nepusz <address@hidden> wrote:
>From what I've been told, UbiGraph can't scale... I think I remember someone stating that 3000 nodes was a high end.
I experimented with Ubigraph some time ago, and it's really nice. There are some problems, though:

- Python's XMLRPC module is not really efficient when it comes to making many XMLRPC calls in a single run. It really takes a lot of time (the author states that he was able to reach 457 API calls per second, which is about the same what I experienced. An API call is required to add a vertex or an edge, so you can calculate how much would it take to show a graph with m nodes and n edges).

- There is a way to link directly with the Ubigraph server, bypassing the XMLRPC layer (which takes a lot of time I guess due to XML parsing), but that's not possible with Ubigraph's free version - although it would be able to handle 740000 API calls per second.

So yes, I think I can confirm that Ubigraph does not scale up to the size of your graph, and it's better to implement your own visualiser.

I also did some experiments with DrL and LGL when I needed a layout for a large graph with ~60K vertices, and I got much better results with DrL (maybe my graph was too dense for LGL). Nevertheless, DrL is still not a realtime layout, so you can only make use of it if your graph is static and you can generate the layout in advance. I think it took about ten minutes to complete on that graph.


@Chris:
Yes, an igraph-Ubigraph bridge would be nice! I think the NetworkX guys have already done something similar, so maybe check that out to see how it scales up to larger graphs.

--
T.


_______________________________________________
igraph-help mailing list
address@hidden
http://lists.nongnu.org/mailman/listinfo/igraph-help


reply via email to

[Prev in Thread] Current Thread [Next in Thread]