Difference between revisions of "Batch Jobs Moab"

From bwHPC Wiki
Jump to: navigation, search
m (Batch Job (PBS) Variables : bwForCluster)
m (Batch Job (Slurm) Variables : bwUniCluster)
Line 586: Line 586:
| Task count (number of PI ranks)
| Task count (number of PI ranks)
* [[Batch_Jobs_-_bwUniCluster_Features|bwUniCluster batch job variables]]
* [[Batch_Jobs_-_bwUniCluster_Features|Additional bwUniCluster Configurations]]

Revision as of 16:49, 8 January 2016

This article is partly outdated and currently under revision!

1 Moab® HPC Workload Manager

1.1 Specification

The Moab Cluster Suite is a cluster workload management package, available from Adaptive Computing, Inc., that integrates the scheduling, managing, monitoring and reporting of cluster workloads. Moab Cluster Suite simplifies and unifies management across one or multiple hardware, operating system, storage, network, license and resource manager environments.
Any kind of calculation on the compute nodes of a bwHPC cluster of tier 2 or 3 requires the user to define calculations as a sequence of commands or single command together with required run time, number of CPU cores and main memory and submit all, i.e., the batch job, to a resource and workload managing software. All bwHPC cluster of tier 2 and 3, including have installed the workload managing software MOAB. Therefore any job submission by the user is to be executed by commands of the MOAB software. MOAB queues and runs user jobs based on fair sharing policies.

1.2 Moab Commands (excerpt)

Some of the most used Moab commands for non-administrators working on a HPC-C4 cluster.

MOAB commands Brief explanation
msub submits a job and queues it in an input queue [msub]
checkjob displays detailed job state information [checkjob]
showq displays information about active, eligible, blocked, and/or recently completed jobs [showq]
showbf shows what resources are available for immediate use [showbf]
showstart returns start time of submitted job or requested resources [showstart]
canceljob cancels a job [canceljob]

1.3 Job Submission : msub

Batch jobs are submitted by using the command msub. The main purpose of the msub command is to specify the resources that are needed to run the job. msub will then queue the batch job. However, starting of batch job depends on availability of the requested resources and the fair sharing value.

1.3.1 msub Command Parameters

The syntax and use of msub can be displayed via:

$ man msub

msub options can be used from the command line or in your job script.

msub Options
Command line Script Purpose
-l resources #MSUB -l resources Defines the resources that are required by the job.

See the description below for this important flag.

-N name #MSUB -N name Gives a user specified name to the job.
-o filename #MSUB -o filename Defines the file-name to be used for the standard output stream of the

batch job. By default the file with defined file name is placed under your
job submit directory. To place under a different location, expand
file name by the relative or absolute path of destination.

-q queue #MSUB -q queue Defines the queue class
-v variable=arg #MSUB -v variable=arg Expands the list of environment variables that are exported to the job
-S Shell #MSUB -S Shell Declares the shell (state path+name, e.g. /bin/bash) that interpret

the job script

-m bea #MSUB -m bea Send email when job begins (b), ends (e) or aborts (a).
-M name@uni.de #MSUB -M name@uni.de Send email to the specified email address "name@uni.de".

For cluster specific msub options, read: msub -l resource_list

The -l option is one of the most important msub options. It is used to specify a number of resource requirements for your job. Multiple resource strings are separated by commas.

msub -l resource_list
resource Purpose
-l nodes=2:ppn=16 Number of nodes and number of processes per node
-l walltime=600
-l walltime=01:30:00
Wall-clock time. Default units are seconds.

HH:MM:SS format is also accepted.

-l pmem=1000mb Maximum amount of physical memory used by any single process of the job.

Allowed units are kb, mb, gb. Be aware that processes are either MPI tasks

memory for all MPI tasks or all threads of the job.
-l advres=res_name Specifies the reservation "res_name" required to run the job.
-l naccesspolicy=policy Specifies how node resources should be accessed, e.g. -l naccesspolicy=singlejob

reserves all requested nodes for the job exclusively.
Attention, if you request nodes=1:ppn=4 together with singlejob you will be
charged for the maximum cores of the node.

Note that all compute nodes do not have SWAP space, thus DO NOT specify '-l vmem' or '-l pvmem' or your jobs will not start. msub -q queues

Queue classes define maximum resources such as walltime, nodes and processes per node and partition of the compute system. Note that queue settings of the bwHPC cluster are not identical, but differ due to their different prerequisites, such as HPC performance, scalability and throughput levels. Details can be found here:

1.3.2 msub Examples

Hint for JUSTUS users: in the following examples instead of singlenode and fat use short and long, respectively! Serial Programs

To submit a serial job that runs the script job.sh and that requires 5000 MB of main memory and 3 hours of wall clock time

a) execute:

$ msub -q singlenode -N test -l nodes=1:ppn=1,walltime=3:00:00,pmem=5000mb   job.sh

or b) add after the initial line of your script job.sh the lines (here with a high memory request):

#MSUB -l nodes=1:ppn=1
#MSUB -l walltime=3:00:00
#MSUB -l pmem=200000mb
#MSUB -N test

and execute the modified script with the command line option -q fat (with -q singlenode maximum pmem=64000mb is possible):

$ msub -q fat job.sh

Note, that msub command line options overrule script options. Multithreaded Programs

Multithreaded programs operate faster than serial programs on CPUs with multiple cores. Moreover, multiple threads of one process share resources such as memory.

For multithreaded programs based on Open Multi-Processing (OpenMP) number of threads are defined by the environment variable OMP_NUM_THREADS. By default this variable is set to 1 (OMP_NUM_THREADS=1).

To submit a batch job called OpenMP_Test that runs a fourfold threaded program omp_executable which requires 6000 MByte of total physical memory and total wall clock time of 3 hours:

  • generate the script job_omp.sh containing the following lines:
#MSUB -l nodes=1:ppn=4
#MSUB -l walltime=3:00:00
#MSUB -l mem=6000mb
#MSUB -v EXECUTABLE=./omp_executable
#MSUB -v MODULE=<placeholder>
#MSUB -N OpenMP_Test

#Usually you should set
export KMP_AFFINITY=compact,1,0
#export KMP_AFFINITY=verbose,compact,1,0  prints messages concerning the supported affinity
#KMP_AFFINITY Description: https://software.intel.com/en-us/node/524790#KMP_AFFINITY_ENVIRONMENT_VARIABLE

module load ${MODULE}
echo "Executable ${EXECUTABLE} running on ${MOAB_PROCCOUNT} cores with ${OMP_NUM_THREADS} threads"
echo $startexe
exec $startexe

Using Intel compiler the environment variable KMP_AFFINITY switches on binding of threads to specific cores and, if necessary, replace <placeholder> with the required modulefile to enable the OpenMP environment and execute the script job_omp.sh adding the queue class singlenode as msub option:

$ msub -q singlenode job_omp.sh

Note, that msub command line options overrule script options, e.g.,

$ msub -l mem=2000mb -q singlenode job_omp.sh

overwrites the script setting of 6000 MByte with 2000 MByte. MPI Parallel Programs

MPI parallel programs run faster than serial programs on multi CPU and multi core systems. N-fold spawned processes of the MPI program, i.e., MPI tasks, run simultaneously and communicate via the Message Passing Interface (MPI) paradigm. MPI tasks do not share memory but can be spawned over different nodes.
Multiple MPI tasks can not be launched by the MPI parallel program itself but via mpirun, e.g. 4 MPI tasks of my_par_program:

$ mpirun -n 4 my_par_program

However, this given command can not be directly included in your msub command for submitting as a batch job to the compute cluster, see above.

Generate a wrapper script job_ompi.sh for OpenMPI containing the following lines:

module load mpi/openmpi/<placeholder_for_version>
# Use when loading OpenMPI in version 1.8.x
mpirun --bind-to core --map-by core -report-bindings my_par_program
# Use when loading OpenMPI in an old version 1.6.x
mpirun -bind-to-core -bycore -report-bindings my_par_program

Attention: Do NOT add mpirun options -n <number_of_processes> or any other option defining processes or nodes, since MOAB instructs mpirun about number of processes and node hostnames. Use ALWAYS the MPI options --bind-to core and --map-by core|socket|node (OpenMPI version 1.8.x). Please type mpirun --help for an explanation of the meaning of the different options of mpirun option --map-by.
Considering 4 OpenMPI tasks on a single node, each requiring 1000 MByte, and running for 1 hour, execute:

$ msub -q singlenode -l nodes=1:ppn=4,pmem=1000mb,walltime=01:00:00 job_ompi.sh

The policy on batch jobs with Intel MPI on bwUniCluster can be found here: Multithreaded + MPI parallel Programs

Multithreaded + MPI parallel programs operate faster than serial programs on multi CPUs with multiple cores. All threads of one process share resources such as memory. On the contrary MPI tasks do not share memory but can be spawned over different nodes.
Multiple MPI tasks using OpenMPI must be launched by the MPI parallel program mpirun. For multithreaded programs based on Open Multi-Processing (OpenMP) number of threads are defined by the environment variable OMP_NUM_THREADS. By default this variable is set to 1 (OMP_NUM_THREADS=1).
For OpenMPI a job-script to submit a batch job called job_ompi_omp.sh that runs a MPI program with 4 tasks and an fivefold threaded program ompi_omp_program requiring 6000 MByte of physical memory per process/thread (using 5 threads per MPI task you will get 5*6000 MByte = 30000 MByte per MPI task) and total wall clock time of 3 hours looks like:

#MSUB -l nodes=2:ppn=10
#MSUB -l walltime=03:00:00
#MSUB -l pmem=6000mb
#MSUB -v MPI_MODULE=mpi/ompi
#MSUB -v MPIRUN_OPTIONS="--bind-to core --map-by socket:PE=5 -report-bindings"
#MSUB -v EXECUTABLE=./ompi_omp_program
#MSUB -N test_ompi_omp
module load ${MPI_MODULE}
echo "${EXECUTABLE} running on ${MOAB_PROCCOUNT} cores with ${TASK_COUNT} MPI-tasks and ${OMP_NUM_THREADS} threads"
startexe="mpirun -n ${TASK_COUNT} ${MPIRUN_OPTIONS} ${EXECUTABLE}"
echo $startexe
exec $startexe

Execute the script job_ompi_omp.sh adding the queue class multinode to your msub command:

$ msub -q multinode job_ompi_omp.sh

With the mpirun option --bind-to core MPI tasks and OpenMP threads are bound to physical cores. With the option --map-by socket:PE=<value> (neighbored) MPI tasks will be attached to different sockets and each MPI task is bound to the (in <value>) specified number of cpus. <value> must be set to ${OMP_NUM_THREADS}.
Old OpenMPI version 1.6.x: With the mpirun option -bind-to-core MPI tasks and OpenMP threads are bound to physical cores. With the option -bysocket (neighbored) MPI tasks will be attached to different sockets and the option -cpus-per-proc <value> binds each MPI task to the (in <value>) specified number of cpus. <value> must be set to ${OMP_NUM_THREADS}.
The option -report-bindings shows the bindings between MPI tasks and physical cores.
The mpirun-options --bind-to core, --map-by socket|...|node:PE=<value> should always be used when running a multithreaded MPI program. (OpenMPI version 1.6.x: The mpirun-options -bind-to-core, -bysocket|-bynode and -cpus-per-proc <value> should always be used when running a multithreaded MPI program.)
The policy on batch jobs with Intel MPI + Multithreading on bwUniCluster can be found here: Chain jobs

A job chain is a sequence of jobs where each job automatically starts its successor. Chain Job handling differs on the bwHPC Clusters. See the cluster-specific pages Interactive Jobs

Policies of interactive batch jobs are cluster specific and can be found here:

1.3.3 Handling job script options and arguments

Job script options and arguments as followed:

$ ./job.sh -n 10

can not be passed while using msub command since those will be interpreted as command line options of job.sh (like $1 = -n, $2 = 10).

Solution A:

Submit a wrapper script, e.g. wrapper.sh:

$ msub -q singlenode wrapper.sh

which simply contains all options and arguments of job.sh. The script wrapper.sh would at least contain the following lines:

./job.sh -n 10

Solution B:

Add after the header of your BASH script job.sh the following lines:

## check if $SCRIPT_FLAGS is "set"
if [ -n "${SCRIPT_FLAGS}" ] ; then
   ## but if positional parameters are already present
   ## we are going to ignore $SCRIPT_FLAGS
   if [ -z "${*}"  ] ; then
      set -- ${SCRIPT_FLAGS}

These lines modify your BASH script to read options and arguments from the environment variable $SCRIPT_FLAGS. Now submit your script job.sh as followed:

$ msub -q singlenode -v SCRIPT_FLAGS='-n 10' job.sh

1.3.4 ForHLR Batch-Jobs

1.3.5 Moab Environment Variables

Once an eligible compute jobs starts on the compute system, MOAB adds the following variables to the job's environment:

MOAB variables
Environment variables Description
MOAB_CLASS Class name
MOAB_GROUP Group name
MOAB_NODECOUNT Number of nodes allocated to job
MOAB_PARTITION Partition name the job is running in
MOAB_PROCCOUNT Number of processors allocated to job
MOAB_SUBMITDIR Directory of job submission
MOAB_USER User name

Most of all scientific programs available for HPC systems are able to extract all essential important environments at their own.
These programs identify the underlying resource management system (TORQUE/Slurm) and use the correct variables.
But a few programs still need 'msub' command line parameters like -np 'number-of-cores...' (example). In this case use TORQUE or Slurm environments only.


  • The MOAB environment variables are for your own convenience only!
  • It's not sure, the contents of the Moab variables are always accurate.
  • Do not use them in your job scripts!
  • Hence use the TORQUE or Slurm environments instead.

1.4 Start time of job or resources : showstart

The following command can be used by any user to displays the estimated start time of a job based a number of analysis types based on historical usage, earliest available reservable resources, and priority based backlog. To show estimated start time of job <job_ID> enter:

$ showstart -e all <job_ID>

Furthermore start time of resource demands, e.g. 16 processes @ 12 h, can be displayed via:

$ showstart -e all 16@12:00:00

For further options of showstart read the manpage of showstart:

$ man showstart

1.5 List of your submitted jobs : showq

The following command displays information about your active, eligible, blocked, and/or recently completed jobs:

$ showq

The summary of your active jobs shows how many jobs of yours are running, how many processors are in use by your jobs and how many nodes are in use by all active jobs. The summary of your active jobs shows how many jobs of yours are running, how many processors are in use by your jobs and how many nodes are in use by all active jobs.

For further options of showq read the manpage of showq:

$ man showq

1.6 Shows free resources : showbf

The following command displays what resources are available for immediate use for the whole partition - for queue "singlenode" - for queue "multinode" - for queue "fat":

$ showbf
$ showbf -c singlenode
$ showbf -c multinode
$ showbf -c fat

For further options of showbf read the manpage of showbf:

$ man showbf

1.7 Detailed job information : checkjob

checkjob <jobID> displays detailed job state information and diagnostic output for the (finished) job of <jobID>:

$ checkjob <jobID>

The returned output for finished job ID uc1.000000 reads:

job uc1.000000

AName: test.sh
State: Completed
Completion Code: 0  Time: Thu Jul 31 16:03:32
Creds:  user:XXXX  group:YYY  account:ZZZ  class:develop
WallTime:   00:01:06 of 00:10:00
SubmitTime: Thu Jul 31 16:02:18
  (Time Queued  Total: 00:00:08  Eligible: 00:03:41)

TemplateSets:  DEFAULT
NodeMatchPolicy: EXACTNODE
Total Requested Tasks: 1

Req[0]  TaskCount: 1  Partition: uc1
Memory >= 4000M  Disk >= 0  Swap >= 0
Dedicated Resources Per Task: PROCS: 1  MEM: 4000M

Allocated Nodes:

SystemID:   uc1
SystemJID:  uc1.000000

IWD:            /pfs/data1/home/ZZZ/YYY/XXX/bwUniCluster
SubmitDir:      /pfs/data1/home/ZZZ/YYY/XXX/bwUniCluster
Executable:     /opt/moab/spool/moab.job.jCLed6

StartCount:     1
Execution Partition:  uc1
Flags:          GLOBALQUEUE
StartPriority:  5321

For further options of checkjob read the manpage of checkjob:

$ man checkjob

1.8 Blocked job information : checkjob -v

$ checkjob -v <jobID>

If your job is blocked do not delete it!

A blocked job has hit a limit and will become idle if resource get free. The "-v (Verbose)" Mode of 'checkjob' also shows a message "BLOCK MSG:" for more details.

BLOCK MSG: job <jobID> violates active SOFT MAXPROC limit of 750 for acct mannheim  partition ALL (Req: 160  InUse: 742) (recorded at last scheduling iteration)

In this case the job has reached the account limit of mannheim while requesting 160 core when 742 were already in use.

1.9 Canceling own jobs : canceljob

canceljob <jobID> cancels the own job with <jobID>.

$ canceljob <jobID>

Note that only own jobs can be cancelled. The command:

$ mjobctl -c <jobID>

has the same effect as canceljob <jobID>.

2 Resource Managers

2.1 TORQUE Resource Manager

The Terascale Open-source Resource and QUEue Manager (TORQUE) is a distributed resource manager providing control over batch jobs and distributed compute nodes. TORQUE can integrate with the Moab Workload Manager to improve overall utilization, scheduling and administration on a HPC-C4 cluster.

2.1.1 Batch Job (PBS) Variables : bwForCluster

At the time a job is launched into execution, TORQUE defines multiple environment variables, which can be used from within the submission script to define the correct workflow of the job.
The most useful of these environment variables are the following:

Environment Brief explanation
PBS_O_WORKDIR Directory where the qsub command is issued
PBS_NODEFILE File that lists the hosts (compute nodes) on which the job is run
PBS_O_WORKDIR Directory where the qsub command is issued
PBS_NODEFILE File that lists the hosts (compute nodes) on which the job is run
PBS_JOBID Unique number PBS assigns to a job
TMPDIR Directory on the scratch (local and fast) disk space that is unique to a job
  • PBS_O_WORKDIR is typically used at the beginning of a script to go to the directory where the qsub command was issued, which is frequently also the directory containing the input data for the job, etc. The typical use is
    inside a submission script.
  • PBS_NODEFILE is typically used to define the environment for the parallel run, for mpirun in particular. Normally, this usage is hidden from users inside a script (e.g. enable_arcus_mpi.sh), which defines the environment for the user.
  • PBS_JOBID is useful to tag job specific files and directories, typically output files or run directories. For instance, the submission script line
    myApp > $PBS_JOBID.out
    runs the application myApp and redirects the standard output to a file whose name is given by the job id. (NB: the job id is a number assigned by Torque and differs from the character string name given to the job in the submission script by the user.)
  • TMPDIR is the name of a scratch disk directory unique to the job. The scratch disk space typically has faster access than the disk space where the user home and data areas reside and benefits applications that have a sustained and large amount of I/O. Such a job normally involves copying the input files to the scratch space, running the application on scratch and copying the results to the submission directory.

2.2 Slurm Resource Manager

The Slurm Resource and Workload Manager (formally known as Simple Linux Utility for Resource Management (SLURM)), or Slurm for short, is a free and open-source job scheduler.

2.2.1 Batch Job (Slurm) Variables : bwUniCluster

Since the work load manager MOAB on bwUniCluster uses the resource manager SLURM, the following environment variables of SLURM are added to your environment once your job has started (only an excerpt of the most important ones).

Environment Brief explanation
SLURM_JOB_CPUS_PER_NODE Number of processes per node dedicated to the job
SLURM_JOB_NODELIST List of nodes dedicated to the job
SLURM_JOB_NUM_NODES Number of nodes dedicated to the job
SLURM_MEM_PER_NODE Memory per node dedicated to the job
SLURM_NPROCS Total number of processes dedicated to the job
SLURM_CLUSTER_NAME Name of the cluster executing the job
SLURM_CPUS_PER_TASK Number of CPUs requested per task
SLURM_JOB_PARTITION Partition/queue running the job
SLURM_JOB_UID User ID of the job's owner
SLURM_JOB_USER User name of the job's owner
SLURM_RESTART_COUNT Number of times job has restarted
SLURM_STEP_NUM_TASKS Task count (number of PI ranks)