Hi All,
I have configured a cluster where each node has a decent graphics card and use it for performing parallel rendering. The caveats of using MS HPC in a GPGPU setting is that you must:
1. Login a console session on all nodes with one user login, lets call him gpu-user.
2. gpu-user must also be the user that that submits jobs otherwise the scheduler will fail because it must match the job submitter to the user that created the console session on each node.
This is troublesome as often the data that we want to render on the cluster is sensitive and only certain users have access to it. Thus using a shared user (gpu-user) for job submission/node console login is non-ideal.
What would be more preferable is that part of the job submission would be to login the job submitters user as the console session on each node, and then also lock the console until the job has completed.
Does anyone know if it is possible to achieve this work flow through the job manager and/or with some scripting?
