Implement a dummy partitioner e.g. doing geographic partitioning either on the node based graph and then try to infer partitions for the edge based graph, or already on the edge based graph with some heuristics. The output should be a mapping from node ids to partition ids.