In this work we present improvements in the speedup and scalability of a distributed fish school simulator. The results were achieved using a new communication strategy for logical processes (LPs) and changing the algorithm of neighbors’ selection that is applied to every fish in each simulation step. In the proposed approach each sender process anticipates future neighbors LP’s data needs. The new proposed strategy reduces communication time by limiting the amount of messages exchanged among LPs. The new neighbors’ selection algorithm was developed with the aim of avoiding unnecessary work. Diminishing the instructions executed by each fish being simulated in each simulation step caused a great reduction in the simulation time.