The configuration Lisa allows users to submit to both "shared" partitions, where they are able to use a subset of the resources of a full node, and "exclusive" partitions, where the full node is allocated for the job. This page explains the partitions available to users, the accounting of using each. Please refer to the user guide for a general introduction on the partition usage and how to submit a job using SLURM.
Lisa partitions
In order to select different hardware where to run, compute nodes are grouped into partitions. Each partition includes a subset of nodes, with a specific maximum wall time and different type of hardware. The table below summarise the partitions available on Lisa which are accessible to users by selecting it with the SLURM option:
#SBATCH --partition=<partition name>
Because of the heterogeneous hardware on the Lisa system, partitions may contain different node types. A specific node type can be requested, within a partition, using the SBATCH
options --constraint
or --gres.
The "Available node features" for each of the partitions on Lisa are listed in the table below.
Partition name | Node type(s) | Available node features | Smallest possible allocation | Max wall time | Notes |
---|---|---|---|---|---|
shared | silver_4110 | --constraint=silver_4110 | 1 core | 120 h (5 days) | |
gold_6130 | --constraint=gold_6130 | ||||
gold_6230R | --partition=shared_52c_384g | ||||
normal | silver_4110 | --constraint=silver_4110 | 1 node | 120 h (5 days) | |
gold_6130 | --constraint=gold_6130 | ||||
fat | gold_6126 | 1 node | 24 h (1 day) | ||
gpu_titanrtx | gold_5118 + TitanRTX | 1 node | 120 h (5 days) | ||
gpu_titanrtx_shared | gold_5118 + TitanRTX | 1/4 of the node, 1 GPU | 120 h (5 days) | ||
gpu_titanrtx_shared_course | gold_5118 + TitanRTX | 48 h (2 days) | Only available during courses | ||
fat_soil_shared | gold_6230 + Titan RTX | 1/4 of the node, 1 GPU | 120 h (5 days) | Contact the service desk to request access to this partition. | |
shared_jupyter | silver_4110 | 2 h | Access available only through the Jupyter hub. |
Accounting
Each account on Lisa has an associated budget. An account can have one or more logins associated with it. Each login has an associated username and password.
In case there are multiple logins associated with one account, these logins all share the same budget.
When you run a job on Lisa, the cost of this job is deducted from your budget. Depending on your affiliation and type of resources, this budget may be without limit:
- UvA and VU users: there is no limit on your CPU budget.
- Other users (NCF, FOM, etc): your CPU budgets have limits.
- All GPU accounts have budget limits.
The unit of accounting is the SBU, which stands for System Billing Unit. The cost of a job in SBUs is equal to the number of wall clock hours the job actually uses (not the estimated time set in the job script), multiplied by the number of cores that are allocated, and a weight factor. For example, you submitted a job script reserving 6 nodes using 16 cores each, with an estimated wall clock time of 4 hours and a weight of 1. Your job is finished after 2 hours and 30 minutes. The cost of your job will then be
6 x 16 x 2.5 x 1.0 = 240 SBU
Since accounting is done based on the actual run time of a job, and not on the estimated (wall clock) run time you put in the job script, there is no penalty in specifying a liberal wall clock time in your job script. The only downside of the latter is that that your job may start later, because it is more difficult for the scheduler to find a slot when to execute the job.
The table below shows the "SBU pricing" of core hours for the various node types.
Node type | # cores per node | Available memory per node | SBUs per 1 hour, full node |
---|---|---|---|
bronze_3104 + GeForce 1080Ti | 12 | 251 GiB | 42.1 |
gold_6130 | 16 | 91 GiB | 16.0 |
silver_4110 | 16 | 91 GiB | 16.0 |
gold_6126 | 48 | 2 TiB | 42.1 |
gold_6230R | 52 | 293.5 GiB | 52.0 |
Shared node accounting
For partitions that allow shared usage (e.g. "shared"), the accounting of the budget is based on the largest resource (cpus/memory/GPUs) that is reserved for the job. This means, for example, that if a job requests half of the memory of a node and a single core, the accounting system will budget the job for half node utilisation. Similarly this applies to the GPU nodes, where requesting half of the GPUs available on the node will budget half of the node SBUs for the duration of the job (see table above for SBUs cost per node).
Costs of inefficient use
You will be charged for all cores in the node(s) that you reserved, regardless of the actual number of cores used by the job/application. So if your application only uses a few (or even one) of the CPU cores of a node, then it makes sense to write a job script that runs multiple instances of this application in parallel, in order to fully utilize the reserved resources.
Getting account and budget information
You can view your account details using
accinfo
This shows information such as the e-mail associated with the account, the initial and remaining budget, and until when the account is valid.
An overview of the SBU consumption can be obtained with
accuse
By default, consumption is shown for the current login, per month, over the last year. Per day usage can be obtained by adding the -d
flag. The start and end of the period shown in the overview can be changed with the -s DD-MM-YYYY
and -e DD-MM-YYYY
flags, respectively. Finally, consumption for a specific account or login can be obtained using -a accountname
and -u username
, respectively.