Distributed visual programming environment: Possibilities and limitations Free

Distributed processing is an essential ingredient for the data processing in the next decade. Surprisingly therefore most state‐of‐the‐art environments lack generic support for distributed processing. It can be realized in a highly portable way using standard tools like RPC and XDR. However the feasibility of such an approach critically depends on the effort the user has to undertake to use these tools and the overhead that is introduced by them. We have developed a visual programming environment which uses an RPC/XDR based client/multiple server approach as execution model and allows, through its sophisticated user‐interface, easy utilization of distributed processing. For a typical heterogenous network, the overhead of communication is measured. The measurements can be described by a linear model: setup times are around 10 ms; per‐8‐byte transfer times, around 10 μs. From these data it can be concluded that distribution of tasks taking less than 50 ms will lead to severe performance degradation. So a coarse grain distribution has to be used. Clearly the influence of the overhead decreases with increasing computation complexity of the tasks distributed. For instrumentation the overhead of this approach is comparable to IEEE‐488 transfer times. Therefore this scheme can readily be used in existing acquisition setups. The combination of distributed processing and visual presentation in this environment can therefore conveniently be exploited in a number of applications within physics.