This research addresses 'componentizing' of existing MPI code. These  MPI components can then be offered as a service from a high-performance server to a user on a single-processor workstation.

We present a method for automated generation of such components with two mechanisms for data transfer: using TCP socket and CORBA sequence. The access to the remote MPI component server is achieved with CORBA in both cases.

Three MPI programs are examined in this work : a matrix multiplication program, an integer sort program, and an FFT program. As a part of the work, comparison is made between the performances offered by socket-based data transfer and CORBA based data transfer between two remote machines.

We show that usage of CORBA interfaced MPI components will help the users attain better performance, even if they access these services from remote servers. Usage of these components with socket-based data transfer as the underlying mechanism will give better performance than CORBA sequence base data transfer. The methodology presented offers data-trasfer mechanism transparency to the user of the components : the user does not have to know how the data transfer mechanism is implemented.