Self-testing of multiprocessor systems with regular diagnostic connections

We propose an approach to organizing self-testing in a multiprocessor system under certain constraints; in particular, we use special diagnostic graphs, and the number of failures does not exceed a certain value $T$. We show that a test experiment where each of $n$ processors is tested by two others lets one determine the state of all except possibly two processors for $T=4$ and except one for $T=3$. The total number of tests does not exceed $2n+2$.