[Bigjob-users] Problems with BigJob example

[Andre Luckow]

Hi Ole,

> so BigJob uses SAGA to launch the agents (e.g., via Globus) but the agents (called sub-jobs?) internally always use (password-less) SSH to launch the 'work units' they pull?
The agent pulls the work-units (aka sub-jobs) and executes them via SSH.


> If that's the case, was that a design decision based on a typical HPC cluster setup (with password-less ssh enabled for MPI)? Is this method sufficient for non-HPC grids, like, e.g., OSG (glidein-wms) as well? Naively, I would assume 'yes', since in an OSG-type of environment, the agent will eventually end-up on an HPC-cluster as well.

Yes, this was the design decision. However, if we hit any issues on
OSG it should be also not too difficult to execute the sub-job without
SSH.

Best,
Andre