[igraph] Critically slow read.graph(), non-linear completion time

[Przemek Grabowicz]

Hi,

I've encountered this when I moved to another workplace. Before I was 
running
$ uname -a
Darwin alice 9.8.0 Darwin Kernel Version 9.8.0: Wed Jul 15 16:55:01 PDT 
2009; root:xnu-1228.15.4~1/RELEASE_I386 i386

Now I'm running
$ uname -a
Linux bob 2.6.24-27-server #1 SMP Wed Jan 27 23:39:33 UTC 2010 x86_64 
GNU/Linux


Now, when I try to run simple command
> system.time(g2 <- read.graph("base/test", format="edgelist", 
directed=TRUE))

I get it done much slower on the new platform. By much slower I mean, 
that the completion time of this task is not growing linearly in case of 
Linux, and it is growing linearly for Darwin. Because of this reading 
big graphs (2.5m nodes) is not possible at all (takes more than a day to 
do this...).
This is independent of formats (I used edgelist and ncol).


There is a way to avoid it with simple trick, by separating file-loading 
and graph-creation, e.g.
gr2<read.table("base/test");
g2<-graph.data.frame(gr2);
Then again completion time is almost growing linearly and it's possible 
to load big graphs.


Do you have any ideas what may be a source of this problem?


Best,
pms.