From owner-chemistry@ccl.net Tue Sep 17 12:02:00 2019 From: "Srijan Chatterjee srijan9x-x-gmail.com" To: CCL Subject: CCL: Optimum CPU use for trajectory analysis Message-Id: <-53847-190917035646-23386-rmnafixDXWamDDn9T0X9hA%server.ccl.net> X-Original-From: "Srijan Chatterjee" Date: Tue, 17 Sep 2019 03:56:44 -0400 Sent to CCL by: "Srijan Chatterjee" [srijan9x()gmail.com] Hi everyone, I am running a gromacs simulation in a cluster with 16 processor on each nodes.I am using -nt option on mdrun command to specify the processor and it is working fine(CPU use is close to 99%). But when i was doing trajectory analysis (gmx traj) i can see that CPU use is only around 1%.As -nt option is not available here then how can i specify CPU usages to get maximum out of my system. Thanks in advance Srijan CPU information CPU info: Vendor: Intel Brand: Intel(R) Xeon(R) CPU E5-2670 0 /./ 2.60GHz No GPU 16 processor each node From owner-chemistry@ccl.net Tue Sep 17 15:46:00 2019 From: "Daniel Roe daniel.r.roe ~ gmail.com" To: CCL Subject: CCL: Optimum CPU use for trajectory analysis Message-Id: <-53848-190917154335-23257-TDIrcxdw4aBP+QeWndZEvw|-|server.ccl.net> X-Original-From: Daniel Roe Content-Type: text/plain; charset="UTF-8" Date: Tue, 17 Sep 2019 15:43:17 -0400 MIME-Version: 1.0 Sent to CCL by: Daniel Roe [daniel.r.roe,gmail.com] Hi, On Tue, Sep 17, 2019 at 1:01 PM Srijan Chatterjee srijan9x-x-gmail.com wrote: > But when i was doing trajectory analysis (gmx traj) i can see that CPU use > is only around 1%.As -nt option is not available here then how can i > specify CPU usages to get maximum out of my system. You may want to try cpptraj (https://github.com/Amber-MD/cpptraj) which has several options for parallelizing analysis (and can read and write Gromacs format trajectories). Most actions during trajectory processing can be parallelized with MPI, and most time-consuming actions are OpenMP-parallelized; both MPI and OpenMP parallelization can be active at the same time. In addition, you can read in ensembles of trajectories (e.g. from REMD simulations) in parallel. For example, if you had a trajectory that was 1000 frames long and you wanted to divide it among 4 MPI processes, you could do something like: mpirun -n 4 cpptraj.MPI -i input.in where input.in contains something like: parm mypdb.pdb trajin mytraj.trr rmsd CA first - -CA out rmsd.agr More details on the parallelism implemented in cpptraj can be found here: https://doi.org/10.1002/jcc.25382 Hope this helps, -Dan > > Thanks in advance > > Srijan > > CPU information > CPU info: > Vendor: Intel > Brand: Intel(R) Xeon(R) CPU E5-2670 0 {} 2.60GHz > No GPU > 16 processor each node> >