Working on UPPMAX

Objectives

  • Let’s get some tips and best practices

Instructor note

  • Approx timing: 14.30-14:45

    • May be expanded if good discussions and interactions

    • Helpers and users can share their solutions

  • Information

  • discussion

$HOME dir and project dir

Quota

  • Disk usage and number of files

  • $HOME has always 32 GB and 300,000 files

  • You typically have one project per project, their size dependent on type of project

  • Compute projects both have disk space (128 GB) and computing time attached to them

  • Check your quota with

$ uquota

Core hours

  • You get core hours only from compute project

  • When they are used you can still get “BONUS” jobs if the resources allow.

  • You can find your current projects (and other projects that you have run jobs in) with the program projinfo.

$ projinfo [<username>]

or

$ projinfo [<project name>]

  • For example, if your project is named NAISS 2017/1-334 you specify naiss2017-1-334

What kind of work will you perform?

@startuml start :login; :Rackham login Node; if (Do CPU/memory intensive work) then (yes) :Use calculation node; if (Do interactive work) then (yes) :$ interactive -A <proj> <options>; else (no) :Make a batch script; :$ sbatch <script>; endif else (no) :Stay on login node and load your module(s): $ module load <software/tool>; :Run tool: $ <toolname> [- options, input, output]; endif :finish; stop @enduml

UPPMAX Cluster overview again!

graph TB Node1 -- interactive --> SubGraph2Flow Node1 -- sbatch --> SubGraph2Flow subgraph "Snowy" SubGraph2Flow(calculation nodes) end thinlinc -- usr-sensXXX + 2FA----> SubGraph1Flow Node1 -- usr-sensXXX + 2FA----> SubGraph1Flow subgraph "Bianca" SubGraph1Flow(Bianca login) -- usr+passwd --> private(private cluster) private -- interactive --> calcB(calculation nodes) private -- sbatch --> calcB end subgraph "Rackham" Node1[Login] -- interactive --> Node2[calculation nodes] Node1 -- sbatch --> Node2 end

Common problems

  • Conda environment clashes with loaded python modules

  • Forgotten environment variables defined in your .bashrc may give unexeptected errors when you run other programs or new versions of a program

  • A full $HOME folder may cause unexpected errors

    • check with uquota

Tip

.bashrc

  • Do not fill ~/.bashrc with too much

    • Avoid loading modules, unless you are totally sure that those will not interact with other things you are doing

    • Typical problem: loading several conflicting conda environments and python modules

  • Good place to have:

    • aliases

    • source files (if needed)

      • these can set some environment variables

  • Example from Björn C

alias lt='ls -lrt'
alias sq='squeue -u $USER'
alias sc='scancel'
alias rm='rm -i'
alias less='less -R'

  • When done, start a new bash session in a new terminal or type bash in the present one.

Conda

  • You cannot mix Conda environments with python software module

  • Conda is very nice for Bianca, because of the ease of install packages without the internet connection.

  • For Rackham we suggest Python virtual environemnts with virtualenv, see Python course

    • Unless the conda environment is straight-forward to install and is well-defined.

Things worked before but do not now

  • Check uquota !

  • Very often a full disk space is the reason

Discussion

  • Any input from other teachers!

Keypoints

  • Use your disk spaces wisely

    • home folder just for general stuff and files needed by several projects

      • always read and write protected for others by default

    • otherside project folder which will more easily become public for others

      • by default available for all project members.

  • Use the computing resources wisely

    • low intensity work on login node

    • high intensity work on compute nodes (core hours are counted)

      • for development use the interactive sessions

      • otherwise make batch jobs!