Requests for Node via Slurm results in CycleCloud spinning up cores instead. RRS feed

  • Question

  • Hello,

    So I have been experimenting with CycleCloud using the Slurm template. Whenever I submit a job that requests 2 nodes (--nodes=2 --ntasks-per-node=1), I am expecting CycleCloud to spin up two actual nodes. Instead, CycleCloud appears to interpret my request for nodes as a request for cores. So it spins up 2 cores (1 node) instead of 2 nodes. How do I configure CycleCloud to see my request from Slurm as 2 nodes instead of 2 cores?



    Monday, June 10, 2019 7:08 PM


  • So, after reaching out to Microsoft Support, it turns out this is not supported yet in CloudCycle. Here is a snippet from their response:

    Hello Eric,


    Thank you for the detailed info. My apologies for the confusion.


    It appears to be, we don't support `-N 2` right now, that won't work properly with our autoscale. If you want two nodes, then we have to use `-n 32` alone(here I am assuming 16 cores for each node). If you want 64 cores, they have to ask for -n 64.


    It looks -N support will be included in the future versions of the CycleCloud.


    Sorry for any confusion.




    Hope that helps anyone else that was confused by this behavior as well.

    • Marked as answer by laforge2001 Monday, June 17, 2019 6:46 PM
    Monday, June 17, 2019 6:46 PM