In case you’re still having this problem, I think the issue here is with FSL’s run_first_all trying to use the Sun Grid Engine (SGE) backend to launch jobs, when it is itself running from an SGE job. I’m guessing on your cluster, the system is configured not to allow compute nodes to submit jobs, which would explain this line in the output:
In which case, you would need to prevent run_first_all from trying to use SGE, and instead run on the local node. Judging from the other relevant threads related to SGE, it might be sufficient to modify your SGE job (the one that invokes 5ttgen) so that the SGE_ROOT environment variable is unset. Maybe by adding this line in the script before 5ttgen is invoked:
unset SGE_ROOT
or prefixing the 5ttgen call with SGE_ROOT="" to clear the variable, like this:
SGE_ROOT="" 5ttgen fsl T1.coreg.nii 5tt_act.mif
Hopefully one of these will work. The idea is to ensure fsl_sub doesn’t detect the presence of an SGE queue - hope that makes sense… More details about this on the FSL wiki.