Hi,

If your graph contains several connected components (subgraphs without links between each other), you can speed these layouts up by setting their SplitGraph property to true. This will arrange each subgraph separately and should be faster since these force-directed algorithms run in polynomial time; running an O(n^3) algorithm three times on 100 nodes is much better than running it once on 300 nodes.

If there are no separate subgraphs but you can divide the graph easily by removing some key links, you can remove them temporarily to allow for SpltGraph to split the graph into several components; setting the IgnoreLayout property of these links should work too. Alternatively, you can run CompositeLayout specifying some smaller subgraphs as custom partitions.

You could also try LayeredLayout with EnforceLinkDirection and StraightenLongLinks properties enabled. Running it with these settings and random graphs of 300-400 nodes in the LayeredLayout sample project takes about 5 seconds on my system.

Finally, if you graphs are trees, use TreeLayout or FractalLayout algorithms - they run in almost linear (O(n)) time. Even if they aren't trees, you might still try using them with some additional processing of non-tree links.

I hope that helps,
Stoyan

In most simple form, CompositeLayout can be set up like this:

private void miLayoutComposite_Click(object sender, EventArgs e)
{
	CompositeLayout cl = new CompositeLayout();
	cl.MultipleGraphsPlacement = MultipleGraphsPlacement.MinimalArea;
	cl.DesiredSubgraphSize = 10;
	cl.MasterLayout = new AnnealLayout();
	cl.SubtreeLayout = new TreeLayout(TreeLayoutType.Radial);
	cl.SubgraphLayout = new CircularLayout();
	cl.Arrange(diagram);
} 

If your graph contains some sort of clusters (e.g. employees in the same company division), you can specify them as custom partitions as shown in the CompositeLayout sample project, so that each cluster's nodes are laid out together. Otherwise the control will try to divide the graph into partitions based on DesiredSubgraphSize and the lengths of paths between nodes.

I hope that helps,
Stoyan