[igraph] graph representation in memory

[Mati Vait]

Hello,
as a school project we are investigating scalability of graph algorithms
on very large undirected graphs. Graphs under question have hundreds of
millions of nodes, billions of edges), but we have run into some memory
issues with igraph because of it's way of holding graph in memory
exhausts ram really fast. We reached this conclusion by looking at
comments on igraph_t in include/igraph.h and testing on cluster with
32GB ram for each node. At the moment we are able to hold in memory a
graph with 100 million nodes and 300 million edges which uses about 27.9
GB of ram (after tweaking igraph_real_t and igraph_integer_t types to
use int instead of double).
Questions:
How difficult it is to implement another way of graph representation
instead of current one? Some loss in work time could be tolerated. 

What code must be reimplemented in order to keep algorithms running?



Some insight and references to specific documentation would be really
helpful.

Thank you for your time,
Mati