Abstract. Wrapping parallel programs or parallel numerical library functions into software components and using them as computation services in serviceoriented programming presents a method of delivering powerful computation capabilities of multi-processor supercomputers to the application developers who may only familiar with their desk-top or hand-held computing environment. These parallel computation services on computer clusters are used as ordinary software components on the desktop programming environment with their internal parallel or distributed characteristics hidden from the users. In order to use the parallel scientific computation applications and libraries as the software components conveniently in the development of new applications, a parallel computation service model and the runtime system that support this model on computer clusters are presented and some design and implementation issues are discussed in this article.