System use
For a more in depth use of the system, we first need to make us more familiar with system infrastructure, specifically resource management and software management.
Resource Management - SLURM
As any other slurm managed cluster you have some different possibilities to execute jobs in it. The main one is to work in batch mode, wich means to submit jobs to the queue. Another option is to work interactively. This can be done through the command compute
, with this command you’ll be provided with an interactive session in x86 nodes dedicated to this job in the ilkinteractivepartition. If you would like to use an interactive session into a64 nodes, you’ll need to use another command. There is not any interactive dedicated node in this partition, so you might need to wait in the queue for resources depending on node availability. salloc -p a64 --mem-per-cpu=600M -t 5:0:0 -c 48 srun -c 48 --pty --preserve-env /bin/bash -i
.
Partitions
ilk
ilkinteractive
a64
qpu
Usefull commands
The best way to extend the information is to ask directly to the cluster with commands as the following ones. batchlim
, sqstat
, scontrol show partition
, sinfo
.
Example with batchlim.
$ batchlim
*******************
Partition limits:
*******************
PartitionName | Default | MaxNodes | MaxTime | Shared | TotalCPUs | TotalNodes | DefMemPerCPU | MaxMemPerNode
---------------|---------|-----------|-------------|-----------|-----------|------------|---------------|----------------
qpu | NO | UNLIMITED | UNLIMITED | EXCLUSIVE | 64 | 1 | 1024MB | UNLIMITED MB
ilk_interactive | NO | UNLIMITED | UNLIMITED | NO | 128 | 2 | ---- | UNLIMITED MB
ilk | YES | UNLIMITED | UNLIMITED | NO | 1280 | 20 | ---- | UNLIMITED MB
a64 | NO | UNLIMITED | UNLIMITED | NO | 768 | 16 | ---- | UNLIMITED MB
*************
QOS limits:
*************
Name Priority GrpTRES MaxTRES MaxWall MaxJobsPU MaxTRESPU MaxSubmit
------------ ---------- ------------- ------------- ----------- --------- ------------- ---------
ilk_short 50 cpu=1408 06:00:00 100 cpu=1408 200
ilk_medium 40 cpu=1408 3-00:00:00 100 cpu=1408 125
ilk_long 30 cpu=640 cpu=640 7-00:00:00 30 cpu=640 30
ilk_ondemand 10 cpu=320 cpu=320 42-00:00:00 10 cpu=320 10
a64_short 50 cpu=768 06:00:00 100 cpu=768 200
a64_medium 40 cpu=768 3-00:00:00 100 cpu=768 125
a64_long 30 cpu=288 cpu=288 7-00:00:00 30 cpu=288 30
a64_ondemand 10 cpu=144 cpu=144 42-00:00:00 10 cpu=144 10
qpu 100 node=1 02:00:00 1 cpu=64 5
ilk_interac+ 100 node=1 08:00:00 2 cpu=8,mem=12+ 2
a64_interac+ 100 node=1 08:00:00 2 cpu=8,mem=50+ 2
...
Software Management - Modules
To be able to use different software options and versions we have configured lmod
package. We use the module
command to ask for specific software.
x86 module tree
This module tree is loaded by default in login nodes and any other node of the system.
a64 module tree
ARM based nodes do not share the same module tree so users need to repoint to the correct module tree using source /etc/profile.d/lmod.sh
.
Qmio Module
To load the python module to interact with the quantum computer you have two options:
module load qmio-run
/module load qmio-tools
This implies that you are going to use other modules compatible with python versions available in the module system. This doesn’t integrate well with per-user enviroments like conda enviroments.pip install qmio
The qmio package is available from pipy so you can install it from there into your enviroment. You can create your own enviroment or install it user-wide withpip install --user qmio
. Check pipy-qmio.
If you use the second method, you have to handle the update by yourself with pip install -U qmio
. We will notify you whenever an update is ready.