User Policy - calab-ntu/gpu-cluster GitHub Wiki

Login Node

  • Fail password for 3 times will be locked for 3 minutes.

  • Maximum number of CPUs: 4

  • Maximum CPU memory: 32 GB

  • Maximum wall time: 1 min

Policy Violating Handling Method

  • Eureka: Users violating the above policies will receive a warning email and the corresponding jobs will be terminated without further notice.
  • Spock: Users exceed the CPU number policies will receive a warning email and the corresponding jobs will be terminated without further notice. Jobs violating the runtime and memory usage policy will be killed automatically.

Computing Node

  • Do not waste resources.
  • Do not ssh to computing nodes unless they have been assigned to you by the queuing system. Consider using the interactive queue (see Running Jobs).
  • Always request the entire node (i.e.,eureka: with ppn=16; spock: with ppn=32 in your job script) even for non-parallel jobs to avoid potential job conflicts.

Storage:

  • Avoid storing simulation data under /home (which has a 1 GB quota) → use /work1 or /project*.

Links

Home