Molecular Simulation Methods with Gromacs

lambda_02lambda_03lambda_04lambda_05lambda_06which each have contents:conf.grogrompp.mdptopol.topVerify that the substitution worked correctly withgrep init-lambda lambda_*/grompp.mdpwhich should show something likelambda_0.2/grompp.mdp:init-lambda = 0.2lambda_0.4/grompp.mdp:init-lambda = 0.4etc. for each ! point. We now need to pre-process each run withcd lambda_00gromppcd ../lambda_01grompp....Check the output for whether these are successful, just to be sure. At this point we areready to run. The total run time will be about 5 minutes on 4 cores on a modern x86(AMD/Intel) CPU per ! point. This means that we can run the jobs sequentially, butthen we’ll have to wait 35 minutes and we’re wasting a big opportunity forparallelization.Instead, we’re going to try to run them in parallel and assume that we have somekind of batch system that we can submit jobs to. Because the system only has 1000particles, scaling beyond 4 cores makes no real gains. Typically, a modern computecluster has 8 core or more cores per node. We will therefore trick our to submit our jobs.Because we use fewer cores than there are in a node, we can use the threaded versionof Gromacs - which doesn’t need MPI (a library and run environment for runningparallel high-performance computing jobs over a network) to run in parallel. In manylocations, Gromacs is installed such that mdrun runs the threaded version, andmdrun_mpi the MPI version, though this may vary.