CASTEP jobs

The topics in this section cover controlling and running remote CASTEP jobs, details of a sample remote run, running jobs in standalone mode, and potential reasons why CASTEP might fail.

Using CASTEP job control

Materials Studio runs CASTEP jobs as background processes on a server. The following tools are provided to setup and control the jobs:

For further information on using job control see the main job control and live updates help topics. Live updates can be requested on the CASTEP Job Control Options dialog.

Running CASTEP jobs in parallel

CASTEP jobs can run in parallel on a server that supports MPI. The CASTEP interface uses the gateway framework of Materials Studio to extract the relevant information for each gateway server. You can select the number of cores to use, up to the maximum number available, as indicated by the text to the right of the Run in parallel on control.

It is recommended to use the number of cores which is commensurate with the number of k-points in the calculation. You can determine the number of k-points by using the Display points... button on the k-points tab of the CASTEP Electronic Options dialog.

CASTEP jobs can run in parallel using hybrid OpenMP/MPI parallelization. CASTEP uses the Materials Studio gateway framework to extract the relevant information for each gateway server. You can select the number of cores to use, up to the maximum number available, as indicated by the Run in parallel on option on the Job Control tab of the CASTEP Calculation dialog. Furthermore, on Linux gateways you can specify the desired number of threads to be used for each MPI process. The number of threads must be commensurate with the number of cores requested and should not exceed the number of cores on each execution host.

This functionality can substantially decrease the amount of memory required for a job (although it could somewhat decrease speed).

Remote CASTEP jobs

CASTEP uses a client-server architecture that allows you use your PC to control calculations of the total energy and electronic properties of a system or geometry optimizations and dynamics simulations of structures running on a remote computer.

This separation of the client user interface from the server system, which runs the calculations, allows you to use a high performance supercomputer from your Windows desktop PC. It also allows you to make use of spare CPU cycles on other desktop PCs.

Execution of server code can be somewhat reconfigured. In particular, for Linux clusters it can be useful to set MaxCoresPerProcessor to the actual number of cores per chip for homogeneous clusters. For example, if a cluster node contains two 6-core chips, the value can be set to 6. This modification can speed up parallel jobs for some architectures.

CASTEP jobs are controlled by input files that are generated by Materials Studio when you start a job. CASTEP writes the results of the calculations to various output files, which are downloaded from the server to the appropriate Materials Studio project when the job is completed.

CASTEP remote jobs run according to the standard sequence of processes described in A sample remote job run with the differences explained in the Sample CASTEP run help topic.

See Also:

Sample CASTEP run
If a remote CASTEP job fails
Running CASTEP in standalone mode
CASTEP keywords
CASTEP file formats