By default GAUSSIAN jobs write temporary working data to the /tmp directory of the compute node. For large GAUSSIAN jobs this /tmp space can fill very quickly causing the machine to slowdown/crash ( note: machine supervisors may pre-empt machine crashes by killing your jobs).
To avoid this situation please direct GAUSSIAN to write temporary working files to your own storage allocation (either NFS or /scratch in future). Details on how to set this up in your GAUSSIAN scripts can be found
at the Gaussian wiki page.