Physnodes - 3) Submitting Jobs to the physics cluster

In order to submit jobs to the physics cluster you need to first load the grid engine module.

Load Grid Engine Tools

-bash-4.1$ module load sge


You will first login to physlogin which allows Linux command line access to the system, which is necessary for the editing programs, compiling and running the code. Usage of physlogin nodes is shared amongst all who are logged in. These systems should not be used for running your code, other than for short test runs.   

Access to the job/batch submission system is through the login nodes. When a submitted job executes, processors on the compute nodes are exclusively made available for the purposes of running the job.   

In order to interact with the job/batch system, the user must first give some indication of the resources they require. At a minimum these include:   

  • how long does the job need to run for  
  • on how many processors to run the job

The default resource allocation for jobs are

Default Resource Allocation
ResourceDefaultMaximum
Time limit-24 hours and 7 days
Memory2GB256GB
Cores1800

Armed with this information, the scheduler is able to dispatch the jobs at some point in the future when the resources become available. A fair-share policy is in operation to guide the scheduler towards allocating resources fairly between users.   

Job submission

Currently the facility is configured with a single general access queue, allowing submission to all available compute resources. Thus, there is no need to specify a queue name in job submissions.   

The general command to submit a job with the qsub command is:

$ qsub [options] script_file_name [--script-args]

where script_file_name is a file containing commands to executed by the batch request.   

For example submission scripts please look at these Job Script FIles.   

Querying queues

The qstat command may be used to display information on the current status of Grid Engine jobs and queues.

List your jobs in the queue
-bash-4.1$ qstat
job-ID  prior   name       user         state submit/start at     queue                          slots ja-task-ID
-----------------------------------------------------------------------------------------------------------------
  10202 0.00000 node_test  abs4         qw    06/02/2014 09:17:30                                    1 1-1000:1
  10203 0.00000 node_test  abs4         qw    06/02/2014 09:17:31                                    1 1-1000:1


ColumnDescription
job-IDA number used to uniquely identify your job within the LSF system. Use this number when you want to halt a job via the qdel command.
riorThe user's current job priority, based upon current and recent cluster utilisation
nameThe job's name, as specified by the job submisison script's -N directive
userThe username of the job owner
stateCurrent job status: r (running), t (in transfer) and qw (queued and waiting)
submit/start atFor waiting jobs: the time the job was submitted. For running the jobs: the time the job started running
queueFor running jobs, the queue and compute node the job is running on
slotsThe number of job slots the job is consuming (1 for serial jobs, greater than 1 for parallel jobs)
ja-task-IDA special field for task arrays


By default, users will only see their jobs in the qstat output. To see all jobs use a username wildcard

List jobs of all users
-bash-4.1$ qstat -u \*
job-ID prior name user state submit/start at queue slots ja-task-ID
-----------------------------------------------------------------------------------------------------------------
10063 0.55500 sh jaw500 r 05/27/2014 13:39:39 elec-cluster@elecnode0.york.ac 1
10064 0.55500 sh jaw500 r 05/27/2014 13:39:39 elec-cluster@elecnode1.york.ac 1
10065 0.55500 sh jaw500 r 05/27/2014 13:39:39 elec-cluster@elecnode2.york.ac 1
10066 0.55500 sh jaw500 r 05/27/2014 13:39:39 elec-cluster@elecnode4.york.ac 1
10067 0.55500 sh jaw500 r 05/27/2014 13:39:39 elec-cluster@elecnode7.york.ac 1
10068 0.55500 sh jaw500 r 05/27/2014 13:39:39 elec-cluster@elecnode6.york.ac 1
10069 0.55500 sh jaw500 r 05/27/2014 13:39:39 elec-cluster@elecnode5.york.ac 1
10070 0.55500 sh jaw500 r 05/27/2014 13:39:39 elec-cluster@elecnode3.york.ac 1

Important switches to qstat are:  


Switch

Action

-help

Prints a list of all options.

-f

Prints full display output of all queues

-g c

Print a 'cluster queue' summary - good for understanding what resources are free, across different queue types

-g t

Print 'traditional' output, i.e. print a line per queue used, rather than a line per job

-u username

Displays all jobs for a particular username.

-j jobidDisplays details about a particular job
-fDisplay in full mode

Job deletion

To delete a job from the queue use the qdel <jobid> command, where jobid is a number referring to the specified job (available from qstat).

Delete a job
-bash-4.1$ qsub slow_job 
Your job 10204 ("slow_job") has been submitted
-bash-4.1$ qstat
job-ID prior name user state submit/start at queue slots ja-task-ID
-----------------------------------------------------------------------------------------------------------------
10204 0.00000 slow_job abs4 qw 06/02/2014 09:26:47 1
-bash-4.1$ qdel 10204
abs4 has registered the job 10204 for deletion
-bash-4.1$ qstat
-bash-4.1$

 

To force action for running jobs use the -f option

A user can delete all their jobs from the batch queues with the -u option 

Information about old jobs

To display a list of recently completed jobs

List recently completed jobs
-bash-4.1$ qstat -s z
job-ID prior name user state submit/start at queue slots ja-task-ID
-----------------------------------------------------------------------------------------------------------------
10201 0.55500 node_test abs4 z 06/02/2014 09:05:02 1 1-1000:1
10202 0.55500 node_test abs4 z 06/02/2014 09:17:30 1 1-1000:1
10203 0.55500 node_test abs4 z 06/02/2014 09:17:31 1 1-1000:1
10204 0.00000 slow_job abs4 z 06/02/2014 09:26:47 1
-bash-4.1$

To see information about completed jobs use the qacct command -o for a user, -d <n> for jobs over the last n days

Information about the system

The qhost command checks the status of the compute nodes

List status of nodes
bash-4.1$ qhost
HOSTNAME ARCH NCPU NSOC NCOR NTHR LOAD MEMTOT MEMUSE SWAPTO SWAPUS
----------------------------------------------------------------------------------------------
global - - - - - - - - - -
ecgberht lx-amd64 2 1 2 2 0.03 3.7G 1.6G 2.0G 65.9M
elecnode0 lx-amd64 32 2 16 32 16.04 62.9G 9.9G 2.0G 0.0
elecnode1 lx-amd64 32 2 16 32 16.03 62.9G 10.1G 2.0G 0.0
elecnode2 lx-amd64 32 2 16 32 15.99 62.9G 7.7G 2.0G 0.0
elecnode3 lx-amd64 32 2 16 32 16.23 62.9G 7.7G 2.0G 0.0
elecnode4 lx-amd64 32 2 16 32 16.16 62.9G 10.1G 2.0G 0.0
elecnode5 lx-amd64 32 2 16 32 16.20 62.9G 10.1G 2.0G 0.0
elecnode6 lx-amd64 32 2 16 32 16.13 62.9G 10.1G 2.0G 0.0
elecnode7 lx-amd64 32 2 16 32 16.09 62.9G 10.1G 2.0G 0.0
rnode0 lx-amd64 32 2 16 32 0.00 62.8G 856.6M 2.0G 0.0
rnode1 lx-amd64 32 2 16 32 1.00 62.9G 834.7M 2.0G 0.0
rnode2 lx-amd64 32 2 16 32 1.00 126.0G 1.2G 2.0G 0.0
rnode3 lx-amd64 32 2 16 32 1.02 62.9G 831.2M 2.0G 0.0
rnode4 lx-amd64 32 2 16 32 1.00 126.0G 1.1G 2.0G 0.0
rnode5 lx-amd64 32 2 16 32 1.01 62.9G 842.1M 2.0G 0.0

FAQ

Why do I get an email saying I am not using the /scratch filestore, when I am?

If you receiving an email about using the wrong directory for job submission and you are sure you are submitting your job from the scratch directory, you probably have edited your file on a Windows PC and transferred it to /scratch.

Editing text files on Windows has the bad effect of inserting additional characters into the file. After you have transferred the file to the login server, use the following command to remove these characters:

dos2unix <job-script>

We do not recommend that you edit your files on a Windows PC and transfer them to the cluster.

My jobs are not running, they just sit in the queue

There are two main causes for your jobs to sit in the queue and do not run (the are in the Pending state:

  1. Use the command :
    qalter -w p job-id


    $ qalter -w v 275566
    verification: found suitable queue(s)

    This means you job script is OK. If you get the message

    verification: no suitable queues

    there is something wrong with your job script, see (2)

  2. The job script is wrong
    1. You may have missed out the runtime parameter "#$ -l h_rt=0:15:00"
    2. You have asked for a resource that is not available, the scheduler will wait for the resource to appear. for example specifying "#$ -l h_vmem=512G" when we have no nodes with 51G Gbytes of memory. Remember total memory requested is slots * memory.
  3. The resources you are requesting are currently in use
    1. You may be requesting for a node with 20 cores and 128GB of memory. This node may be being used by another job, the node may be ether fully utilised, or only one slot may be active.
    2. The queue suitable for the job is not currently enabled. For example a job requiring 48 hours to run will only start on the weekend queue, which is enabled on Friday evening.
  4. Use the command 'qstat -j <jobid>' to display why the job is currently running.

What resources did my job use

In order to tune your job submission parameters use the '-me' directive to inform you of the resources used:

Job end email
Job 754967 (mpi-16_job) Complete
 User             = abs4
 Queue            = its-4hour@rnode2.york.ac.uk
 Host             = rnode2.york.ac.uk
 Start Time       = 02/18/2015 10:41:21
 End Time         = 02/18/2015 10:41:27
 User Time        = 00:00:01
 System Time      = 00:00:01
 Wallclock Time   = 00:00:06
 CPU              = 00:00:03
 Max vmem         = 69.367M
 Exit Status      = 0

Here we can see the job used 3 seconds of CPU time an 69 MB of memory.

More information can be found by using the "qacct" command:

Resources used by a terminated job
abs4@login2.york.ac.uk$ qacct -j 830402
==============================================================
qname        yarcc-testing       
hostname     tnode0.york.ac.uk   
group        csrv                
owner        abs4                
project      NONE                
department   defaultdepartment   
jobname      bad.job             
jobnumber    830402              
taskid       undefined
account      sge                 
priority     0                   
qsub_time    Tue Nov 22 08:25:59 2016
start_time   Tue Nov 22 08:28:04 2016
end_time     Tue Nov 22 08:28:04 2016
granted_pe   NONE                
slots        1                   
failed       0    
exit_status  0                   
ru_wallclock 0s
ru_utime     0.031s
ru_stime     0.041s
ru_maxrss    5.395KB
ru_ixrss     0.000B
ru_ismrss    0.000B
ru_idrss     0.000B
ru_isrss     0.000B
ru_minflt    7481                
ru_majflt    0                   
ru_nswap     0                   
ru_inblock   8                   
ru_oublock   24                  
ru_msgsnd    0                   
ru_msgrcv    0                   
ru_nsignals  0                   
ru_nvcsw     93                  
ru_nivcsw    19                  
cpu          0.071s
mem          0.000GBs
io           0.000GB
iow          0.000s
maxvmem      0.000B
arid         undefined
ar_sub_time  undefined
category     -U special14day,iggi-users,special3day,testusers -u abs4 -l h_rt=120,h_vmem=1G