This page was translated by an AI (LLM) with a cursory human check and is awaiting full review.
Monitoring of computing hours
Hour deductions
On Jean Zay, computing hour allocations are made per project, differentiating between CPU and GPU computing hour deductions.
The deduction of hours consumed by your jobs is based on:
- the number of physical cores reserved × Elapsed time for a CPU-type job
- or the number of GPUs reserved × Elapsed time for a GPU-type job.
For a GPU-type job, the number of reserved GPUs
taken into account for the job's accounting is not necessarily
equal to the number of GPUs requested via the Slurm directive
#SBATCH --gres=gpu:...
Indeed, an adjustment may be made
depending on the number of CPUs that is also requested for the job. For example, requesting a single GPU from a node (i.e. 1/4 or 1/8 of the node's GPUs depending on the node type) via #SBATCH --gres=gpu:1 and half
of the node's CPUs results in the reservation of half of the node's resources. The job's accounting is then calculated based on a reservation of half of the GPUs (as if you had requested
#SBATCH --gres=gpu:2 or #SBATCH --gres=gpu:4 depending on the node type).
Note that during execution in exclusive mode (option --exclusive
of the command sbatch or Slurm directive #SBATCH --exclusive),
you reserve the compute nodes in full:
- either 40 physical cores for the CPU partitions;
- either 40 physical cores and 4 GPUs for the default GPU partition;
- either 24 physical cores and 8 GPUs for the
gpu_p2partition. - either 64 physical cores and 8 GPUs for the
gpu_p5partition.
The deduction will then be based on:
- the number of reserved nodes × 40 cores × Elapsed time for a CPU-type job;
- the number of reserved nodes × 4 GPUs × Elapsed time for a GPU-type job in the default GPU partition;
- the number of reserved nodes × 8 GPUs × Elapsed time for a GPU-type job in the
gpu_p2andgpu_p5partitions.
All executions requesting one or more compute nodes are automatically done in exclusive mode and are therefore billed based on the number of whole nodes reserved.
You can view the allocations for your project using the following command:
idracctAdditionally, we recommend that you consult our documentation regarding automatic hour regulation which governs the use of computing hours.
Hour allocations
Several scenarios are possible depending on whether your login is associated with a single project or multiple projects and whether the project(s) have CPU, GPU V100, GPU A100, GPU H100 hours, or at least two of the four:
- If your single project has only CPU hours or GPU hours of a single type (V100, A100 or H100), you do not need to specify the hour allocation on which the hours consumed by your computations should be deducted.
- However, if your single project has both CPU
and GPU hours or GPU hours of multiple types (among V100, A100 and H100) or if your login is associated with multiple projects, you must
imperatively specify the allocation on which the hours consumed by your computations should be deducted. Without this
specification, you risk deducting your consumed hours from a different project than the one you wish to work on.
You must therefore obligatorily indicate the correct project (
my_project) and the type of hours to use (cpu,v100,a100orh100) :- either via the addition of the option
--account=...to the commandsbatch:
sbatch ... --account=my_project@cpu job_cpu.slurm # for CPU hourssbatch ... --account=my_project@v100 job_gpu.slurm # for GPU V100 hourssbatch ... --account=my_project@a100 job_gpu.slurm # for GPU A100 hourssbatch ... --account=my_project@h100 job_gpu.slurm # for GPU H100 hours- or via the Slurm directive
#SBATCH --account=...in the submission script header:
#SBATCH --account=my_project@cpu#SBATCH --account=my_project@v100#SBATCH --account=my_project@a100#SBATCH --account=my_project@h100 - either via the addition of the option
- The type of hours indicated (
cpu,v100,a100orh100) must be in accordance with the partition and the QoS chosen for the job execution (#SBATCH --partition=cpu_...and#SBATCH --qos=qos_cpu-...) or (#SBATCH --partition=gpu_...and#SBATCH --qos=qos_gpu-...). For more information, you can consult our documentation detailing the partitions and QoS (CPU and GPU). - The environment variable
IDRPROJreferences the active project to which you are currently attached. Its value (echo $IDRPROJ) can therefore be used to populate the stringmy_project. - To find out all possible values to populate the string
my_projectand which reference your different projects, you can use the IDRIS commandidrproj(seeidrproj -h). For example, for a login associated with two projectsprj1andprj2:
In this case,idrprojAvailable projects:
------------------------------------------------------------------------
prj1 (999999) [default][active] prj2 (999998)my_projectcan be set toprj1orprj2.
The idracct command
A simple call to the command idracct will indicate, for each of
your projects, the CPU and/or GPU hour allocation, as well as the hourly consumption of each member.
Here is an example with the project 999999 grouping 2 users login1 and login2 : here the 1000000 CPU hours and the 40000 GPU V100 hours are allocated for the period from 01 November 2019 to 31 October 2020 :
idracctLast update on 15-11-2019 08:00:01
################################################################################ PROJECT 999999 ON JEAN-ZAY################################################################################
================================================================================ CPU (from 01-11-2019 to 31-10-2020) Allocation : 1000000 h.cpu================================================================================
Account Owner Consumption Number of jobs Refund -------- ------------------------------ ------------- ---------- ------------- login1 Name1_Surname1 42809.00 10 - login2 Name2_Surname2 12014.00 50 - -------- ------------------------------ ------------- ---------- ------------- Totals 54823.00 60 0.00 Usage 0.05%
================================================================================ V100 (from 01-11-2019 to 31-10-2020) Allocation : 40000 h.gpu================================================================================
Account Owner Consumption Number of jobs Refund -------- ------------------------------ ------------- ---------- ------------- login1 Name1_Surname1 29078.83 248 - login2 Name2_Surname2 2706.31 163 - -------- ------------------------------ ------------- ---------- ------------- Totals 31785.14 411 0.00 Usage 79.46%The command idracct has various options. To view them,
use the option -h or --help. Here are the most useful ones:
- The option
-a(or--accounting) allows you to specify the type of hours desired in the command output: CPU, V100, A100 or H100. Be sure to use uppercase letters here!
idracct -a V100Last update on 15-11-2019 08:00:01################################################################################
PROJECT 999999 ON JEAN-ZAY################################################################################================================================================================ V100 (from 01-11-2019 to 31-10-2020) Allocation : 40000 h.gpu================================================================================
Account Owner Consumption Number of jobs Refund -------- ------------------------------ ------------- ---------- ------------- login1 Name1_Surname1 29078.83 248 - login2 Name2_Surname2 2706.31 163 - -------- ------------------------------ ------------- ---------- ------------- Totals 31785.14 411 0.00 Usage 79.46%- the option
-s(or--summary) allows you to obtain the information in a summarised format:
idracct -sLast update on 15-11-2019 08:00:01Project Category Start End Budget Unit Consumption Number of jobs Refund Usage------ --------- ---------- ---------- -------- ----- ------------ --------- ------------- -------999999 CPU 01-11-2019 31-10-2020 1000000 h.cpu 54823.00 60 0.00 0.05%999999 V100 01-11-2019 31-10-2020 40000 h.gpu 31785.14 411 0.00 79.46%