When using the HPC systems, all programs, data processing, mathematical models, and simulations are run by submitting jobs to a batch scheduling system (known as PBS). Unlike personal or workstation computers these jobs are not interactive. Unlike the personal or workstation computers, programs are not started by clicking the launch or run button. The batch scheduling system then runs them on the appropriate system without any user interaction (unless specified) at a time where there are no schedule conflicts.
Users are charged for the jobs submitted to the batch system a number of hours equal to the number of processors (CPUs) used by the program times the length of time the program runs (walltime). This amount is subtracted from the monthly allocated CPU-Wallclock-hours that each group is given. The amount of hours depends on the group and the queue type selected. All group members running jobs throughout the month use the allocated CPU-Wallclock-hours when running jobs on the HPC systems.
Please note that these allocations are not automatically added upon account creation and must be manually added by the HPC consultant team. Please email them at firstname.lastname@example.org if you find your account has no hours allocated.
High Priority cpu-wallclock hours are allocation to 'Buy-In' users in proportion to the nodes/cpus purchased. Each cpu/core purchased will receive a high priority allocation of 30 days * 24 hours * 1 core = 720 cpu-wallclock hours (based on an average of 30 days per month).
Standard priority is lower than high-priority, but higher than windfall.
Each research group will receive a monthly standard allocation of 12,000 cpu-wallclock hours per user for use on the SGI Altix UV, SGI Altix ICE 8400 and IBM iDataPlex HTC systems. Each user will only be counted once and the 12,000 cpu-wallclock hour allocation will be added to their primary group allocation.
All monthly allocations will be refreshed on the first day of each month. Unused allocations will not be rolled forward into the next month.
All users can also use an unlimited number of cpu-wallclock hours in the lowest priority queue known as 'windfall'. There is no allocation of windfall, lowest priority cpu-wallclock hours. However, windfall jobs run the risk of being preempted (killed and requeued) if there are standard priority or high-priority jobs that are submitted and there are no idle processors available to run those jobs.