The degree of parallelism in generalized stochastic network

We propose a novel concept of parallelism degree for generalized stochastic networks. This concept could be used in design of real-time parallel computing systems. It characterizes the maximal load which does not lead to queue emergence. In the case when arc duration distributed according to arbitrary bounded distributions the parallelism degree estimates the minimum number of processors in the network at which no queues emerges on the network arcs. We also developed a method for finding this parameter.