Skip to content

Interactive nodes

  • Interactive nodes only exist on Spirit/Spiritx CPU cluster
  • Hal GPU cluster has only an ssh gateway
  • interactive work must be done on cluster via slurm (GPU Hardware cannot be shared easily)

On Spirit/Spiritx, Interactive nodes are dedicated to compilation and short run with low memory requirements. cpu and memory is shared between all users.

Restrictions on Spirit/Spiritx:

  • internal users :
    • 60 minutes cpu time maximum per process
    • 8Gb of memory
    • 300 process
    • 32Gb vmem 16Gb mem for all ssh sessions initied for a same user
  • external users :
    • 30 minutes cpu time maximum per process
    • 4Gb of memory
    • 300 process
    • 32Gb vmem 16Gb mem for all ssh sessions initied for a same user

If memory requirement or cpu time is over those limit , your process will be automatically terminated.

If you reach the 300 process limits , you obtain cannot fork message.

Monitoring

Memory usage

To know how much memory your program is using : top -u $USER 

top - 08:05:24 up 70 days, 16 min, 102 users,  load average: 0.19, 0.26, 0.17  
Tasks: 1606 total,   1 running, 1570 sleeping,  25 stopped,  10 zombie  
Cpu(s):  0.0%us,  0.1%sy,  0.0%ni, 99.9%id,  0.0%wa,  0.0%hi,  0.0%si,  0.0%st  
Mem:  132123472k total, 33127768k used, 98995704k free,   754332k buffers  
Swap: 102399996k total,  5823616k used, 96576380k free, 10561812k cached  

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND  
64256 xxxxxx    20   0 7446m 566m 134m T  0.0  0.4   0:22.04 MATLAB

If your requirement is over those limits you have to launch an Interactive Job through the job manager