Abstract : In this paper, we show a parallel programming language Sushi, focusing on the features designed to improve productivity of programmers using parallel computers or a cluster of single CPU computers. A program for such parallel computers includes how an algorithm is described, and how to tune this. A sushi program consists of a computation part (a set of process descriptions and their connections by streams) and an annotation part. Processes can work concurrently communicating each other by streams. An annotation part does not affect the algorithm of a computation part but controls process migration and the scheduling priority of processes. With the annotation, programmers can use parallel computers flexibly and effectively.