We are concerned with the performance analysis of finite-source networks with users retrials (repeated calls) and two heterogeneous servers classes using Markov Chains Model. Emphasis is done on comparison of two servers allocation disciplines, namely the Random Server discipline and the Fastest Free Server discipline. It is known that an analytical solution of the corresponding retrial queueing model is difficult even in the homogeneous servers case. Hence, this paper aims at developing, for each allocation discipline, an algorithm for computing the corresponding infinitesimal generator which allows a direct numerical calculation of steady state probabilities. Moreover, we developed formulas of the main stationary performance measures of the network and we've compared the two disciplines through numerical examples.