How does head node pass the job to the computational node? RRS feed

  • Question

  • How does head node pass the job to the computational node? With SQL profiler I checked the behavior of computational node during submission and execution. SQL profiler showed that only head node had accessed to the DB. Now I need to understand the low level communication between head node and computational node in these two cases:
    1. assigning job to computational node,
    2. passing some information to computational node.
    Wednesday, March 15, 2017 11:52 AM


  • Hi,

      Could you tell us what scenario you want to accomplish so that we can provide more accurate information?

      And here are some simple answers to your questions:

    1. Scheduler Service running on the headnode will query the DB, getting information about Node Resource and Job Queue. And according to scheduler policy, node resource will be assigned to queued job, after a job get node resource, the job will go to running state. And then the tasks in the job will be dispatched to the compute node (With Job and Task information). And then the task will get executed on the compute node

    2. Job/Task information are passed to the NodeManager service running on compute node from scheduler service. And then the result will return back to the scheduler from the nodeManager when the task finishes. in HPC Pack 2012 R2, the communcaition between headnode and windows computenode is through .net remoting channel while the communication between headnode and linux computenode is through REST call

    Qiufang Shi

    • Marked as answer by Artem Azaryan Thursday, March 16, 2017 6:28 AM
    Wednesday, March 15, 2017 2:39 PM