Question

I have a large graph (30k vertices, 250m edges) and using boost graph library adjacency list (I tried both vecs and lists) consumes more than 25gb. since it is not very easy to get a pc with more than 16gb ram, what do you recommend to decrease memory usage ?

Answer 1

I’ve had the same problem with a verification technique using graphs I’m working on.
If you will work with a huge amount of data (vertices and also edges) you should use a fine tuned data structure created by yourself.

If you have a lot of edges, you should consider using an Adjacency Matrix.
The edges have weights, so you could use something like this:

vector<int> vertices;
vector<vector<int> > edges;

You could use the optimized vector<vector<bool> > that uses only one bit to represent each edge, if the edges didn’t have a property.

Although the adjacency matrix is fast when you want to check for an edge or add one, it’s not good to iterate over edges.

I can say that, using the same properties in BGL and in my own graph, my graph is way smaller.