One of important tasks of cluster implementation is performance evaluation and optimal number of working nodes selected for different tasks. Important facility of cluster is possibility to use parallel algorithm for calculations instead serial. The speedup of using parallel processing on a problem, versus using only processor (core) was defined by G.Amdahl's Law. Practically not full task is possible to parallelize. In any case there is some serial overhead. Percentage of overhead depends on amount of each task. In this paper some measurements on interconnections within cluster will be analyzed. The paper will be interesting for peoples who are optimizing parallel job workflow.