Parallel external sorting for CUDA-enabled GPUs with load balancing and low transfer overhead