CONN grid help

Using CONN in a computer cluster environment

CONN grid computing options allow you to process your subjects in parallel, with each subject, or each of several smaller groups of subjects, being processed independently by a different computer node. This can significantly increase the speed of your analyses, allowing you to process hundreds of subjects in the time it would take to process just one or a few subjects. Configuration/Installation: To use this functionality you need to have access to a computer that is part of a distributed cluster environment. Using or installing CONN on a cluster environment does not require administrator privileges. Each node in your cluster simply needs to have access to the folders containing: * option 1) SPM, CONN, and Matlab (and associated Matlab licenses) * option 2) standalone CONN, and Matlab Runtime (MCR, freely available; no licenses required) * as well as the data folders containing your conn project and data CONN supports natively the following cluster computing job schedulers: * Grid Engine (Sun/Oracle Grid Engine, Open Grid Scheduler, or compatible system) * PBS/Torque (Portable Batch System, or compatible) * LFS (Platform Load Sharing Facility, or compatible) * Slurm (Simple Linux Utility for Resource Management, or compatible) CONN will work out-of-the-box in many institutional clusters simply by selecting the appropriate job scheduler that handles job submission in your cluster environment. In addition, profiles for other schedulers, or system-specific settings (e.g to increase your system default walltime settings, enter optional project or account ids, etc.), can be easily created/edited in CONN Tools.Cluster/HPCsettings if necessary. To get started simply install CONN on a computer that is part of your distributed cluster environment. Then launch CONN GUI interactively and select Tools.Cluster/HPCsettings. There you can select and test your system job scheduler. If in doubt simply try one of the four default configurations (GridEngine, PBS, LSF, or SLURM) and click on 'Test profile' (each test may take up to a few minutes). If you see a 'failed' message, click 'Cancel' and try a different profile. If you see a 'Test finished correctly' message, select this as your default profile and save your profile configuration options for future sessions and/or for other users. This process needs to be performed just once. Usage: When running any of CONN's analysis steps (preprocessing, Setup, Denoising, and First-level) you can select between running them locally on your current computer or running them in parallel. If using CONN's GUI, before pressing the 'Start' button to start your desired analysis step, simply change the field that reads 'local processing' to 'distributed processing'. You will then be prompted to select the number of jobs that you want to submit (typically you may divide your analyses into as many as one job per subject; keep in mind that your cluster environment may have some recommendations or limitations regarding the maximum number of jobs that may be run simultaneously, or the maximum-time that individual jobs will be allowed to run). CONN will then automatically: a) break down your analysis into the selected number of jobs b) submit these jobs to your cluster (using your default cluster configuration options) c) track these jobs progress and let you re-submit any failed jobs if necessary and d) merge the results back into your conn project when they are finished If using CONN's batch processing options, simply enter the desired number of jobs/nodes in the batch.parallel.N field as part of your conn_batch fields. See help conn_batch for additional details
see also CONN_JOBMANAGER description of cluster/HPC configuration options in CONN