• Q1: You need to make sure that the MPI processes execute on distinct nodes of Lonestar cluster. This additional precaution is needed because Lonestar nodes are quadcore machines. So the batch system usually schedules consequetive processes on cores of same processor. You can run MPI processes on seperate nodes by gaming the batch system. Specify more number of processors than actually used when scheduling jobs. Also look at -map and -npmin parameters of IMB benchmark, to achieve this effect.