Re: [Boost-users] [Graph] Distributed edge insertion

On 8/24/15 8:13pm, Sensei wrote:

Dear all,

I am finding difficulties in understanding how distributed graphs work. My graph is simple, a string-valued property, with some edges, however, right now I have only one edge... if it works.

Since each process will add its own nodes, is it possible to avoid communication when adding nodes, given that each node owns its own vertices, and no overlap is possible between processes? This would save some communication avoiding re-distribution of nodes.

Can I add edges concurrently? Meaning, each process adds some edges. I've only seen examples with root adding edges, but that's limiting.

My last question is maybe easy. I'm adding an edge that connects two nodes, owned by different processes. Doing this results in a crash. It seems I cannot access properties with operator[] when traversing processes.

This issue seems to be solved by avoiding operator[], and using a property map: property_map::type properties = get(&str::s_, g); synchronize(g); BGL_FORALL_EDGES(e, g, dgraph) { std::cout << "E @ " << comm.rank() << " " << get(properties, boost::source(e, g)) << " -> " << get(properties, boost::target(e, g)) << " owner " << e.owner() << " srccpu " << e.source_processor << " dstcpu " << e.target_processor << std::endl; } However, I am still trying to figure out the answers to the first two questions: - is it possible to add edges from different processes concurrently? - is it possible to partition nodes a priori? Of course my code works right now, but maybe what I'm doing isn't kosher, that's why I'm asking if it's possible (aka "safe"). Thanks!