Multiprogrammation of parallel applications on the PAROS operating system kernel

Parallel machines based on the distributed memory paradigm are basically single-user batch systems attached to a host running a general-purpose operating system. These systems are rather static with regard to resource allocation. The PAROS microkernel has been designed to provide dynamic support for general-purpose parallel applications and operating systems. The ability to efficiently support many users and environments concurrently has been one of the objectives of the design. In this paper, we present a global evaluation of the performance of PAROS obtained when executing both single and multiprogrammed applications. We have concentrated our interest on the system behavior for multiprogrammed parallel applications, observing CPU and links utilization, together with the application's throughput.<
>