Executing on top of the native operating system, the PVM programmming interface clearly induces additional overheads. The slower the application processor, the more time is spent in setting up the communication between processors. Compared to Parix, PVM needs more bookkeeping for the management of communication buffers, for any necessary data conversions and for additional system calls.
The performance losses are especially pronounced in the results obtained on the large transputer system (Table 1). Here, the PVM implementation took 79 seconds on a 512 node-system, while the Parix implementation computed the same solution in only 51 seconds.
Due to unpredictable execution times, time-critical interactive applications should better not be run on workstation clusters. Different processor capabilities may induce load imbalances and the system load might change dynamically during the program runtime. While the different processor speeds did not affect the performance of our PVM implementations (which includes a dynamical load balancing strategy), we still observed widely varying execution times from one execution run to the next.