Since the edges in each shard, on each node, are sorted by vertex id, as the note says, when we iterate over all intervals we have contiguous memory access over each shard. This is useful as it will minimize the cost for accessing the data we need which is important when these large graphs are stored in a distributed environment.