modular-jobs.rst 5.12 KB
Newer Older
Benedikt von St. Vieth's avatar
Benedikt von St. Vieth committed
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139
.. include:: system.rst

.. _het_modular_jobs:

Heterogeneous and Cross-Module Jobs
===================================

.. _het_modular_jobs_overview:

Overview
--------

.. _het_modular_jobs_slurm:

Slurm Support for Heterogeneous Jobs
------------------------------------
For detailed information about Slurm, please take a look on the :ref:`Quick Introduction <quickintro>` and :ref:`Batch system <batchsystem>` page.

With Slurm 17.11 support for Heterogeneous Jobs was introduced. This allows to spawn a job across multiple partitions of a cluster, and across different Modules of our Supercomputers. See the official Slurm documentation (SlurmHetJob_) for additional informations on this feature.

.. _SlurmHetJob: https://slurm.schedmd.com/heterogeneous_jobs.html

**salloc/srun**

.. code-block:: none

  salloc -A <budget account> -p <batch, ...> : -p <booster, ...> [ : -p <booster, ...> ]
  srun ./prog1 : ./prog2 [ : ./progN ]

**sbatch**

.. code-block:: none

  #!/bin/bash
  #SBATCH -A zam
  #SBATCH -p <batch, ...>
  #SBATCH packjob
  #SBATCH -p <booster, ...>

  srun ./prog1 : ./prog2


.. _het_modular_jobs_software:

Loading Software in a Heterogeneous Environment
-----------------------------------------------
Executing applications in a modular environment, especially when different Modules have different architectures or the dependencies of programs are not uniform, can be a challenging tasks.

**Uniform Architecture and Dependencies**

  As long as the Architecture of the given modules are uniform and there are not mutually exclusive dependencies for the binaries that are going to be executed, one can rely on the ``module`` command. Take a look on the :ref:`Quick Introduction <quickintro>` if ``module`` is new for you.

  .. code-block:: none

     #!/bin/bash -x
     #SBATCH ...
     module load [...]
     srun ./prog1 : ./prog2

**Non Uniform Architectures and Mutual Exclusive Dependencies**

  A tool called ``xenv`` was implement to ease the task of loading modules for heterogeneous jobs. For details on supported command line arguments, execute ``xenv -h`` on the given system.

  .. code-block:: none

      srun --account=<budget account> --partition=<batch, ...> xenv -L intel-para IMB-1 : --partition=<knl, ...> xenv -L Architecture/KNL intel-para IMB-1

.. ifconfig:: system_name == 'jureca'

.. _het_modular_jobs_mpi_bridges:

MPI Traffic Across Modules
--------------------------
When the nodes of a job belong to different interconnects and MPI communication is used, bridging has to take place. To support this workflow, e.g. run a job on a Cluster with Infiniband and a Booster with OmniPath, a Gateway Daemon (psgwd, ParaStation Gateway Daemon) was implemented that takes care of moving packages across fabrics.

To request gateway nodes for a job the mandatory option --gw_num has to be specified. In addition, communication with the psgwd has to be ensured via loading the software module **pscom-gateway** either via ``xenv`` or the ``module`` command.

To start an interactive pack job using two gateway nodes the following command can be used:

.. code-block:: none

  export PSP_GATEWAY=2
  srun -A <budget account> -p <cluster, ...> --gw_num=2 xenv -L pscom-gateway ./prog1 : -p <booster, ...> xenv -L pscom-gateway ./prog2

Where ``PSP_GATEWAY=2`` ensures that the gateway protocol is used, **not** that two gateways are used!

For debugging purposes, and to make sure the gateways are used, you might use

.. code-block:: none

  export PSP_DEBUG=3

You should see output like

.. code-block:: none

  <PSP:r0000003:CONNECT (192.168.12.34,26708,0x2,r0000003) to (192.168.12.41,29538,0x2,r0000004) via gw>
  <PSP:r0000004:ACCEPT  (192.168.12.34,26708,0x2,r0000003) to (192.168.12.41,29538,0x2,r0000004) via gw>

PSGWD
~~~~~
The psgw plugin for the ParaStation management daemon extends the Slurm commands salloc, srun and sbatch with the following options:

.. code-block:: none

  --gw_file=path Path to the gateway routing file
  --gw_plugin=string Name of the route plugin
  --gw_num=number Number of gateway nodes

A routing file will be generated in $HOME/psgw-route-$JOBID. The routing file is
automatically removed when the allocation is revoked. With the option --gw_file an
alternative location using an absolute path for the routing file can be specified:

.. code-block:: none

  srun --gw_file=/home-fs/rauh/route-file --gw_num=2 -N 1 hostname : -N 2 hostname

The route plugin can be changed using the --gw_plugin option. Currently only the
default plugin “plugin01” is available.

.. code-block:: none

  srun --gw_plugin=plugin01 --gw_num=2 -N 1 hostname : -N 2 hostname

If more gateways were requested than available the slurmctld prologue will fail for
interactive jobs

.. code-block:: none

  srun --gw_num=3 -N 1 hostname : -N 2 hostname
  srun: psgw: requesting 3 gateway nodes
  srun: job 158553 queued and waiting for resources
  srun: job 158553 has been allocated resources
  srun: PrologSlurmctld failed, job killed
  srun: Force Terminated job 158553
  srun: error: Job allocation 158553 has been revoked

If batch jobs run out of gateway resources they will be re-queued and have to wait for
10 minutes before becoming eligible to start again.