Now the source code of `spark-bench` uses single scala time function to evaluate the execution time. The time of `querytime` is incorrect if we just call the sql function which will return right now.  