Main   News   Research   Members   Courses   Notes   Publications   Opportunities   Resources   Q-Chem   PSI   Group Area

Search the Sherrill Group

The following instructions are written for the Sherrill group and collaborators. The job classes mentioned here are only accessible to users in group sherrill.

The SP2 has a batch queue system called LoadLeveler which it uses to schedule when jobs will run on what nodes. It is necessary to tell LoadLeveler what kind of calculation you are running, in addition to the particular script or program to execute. Since each node has only 4 processors, and only limited amounts of memory and disk, it is critical to ensure that a situation never arises in which calculations are scheduled which use more processors or memory or disk than there actually is on a node. It is really up to the user to provide the information the LoadLeveler program needs to avoid catastrophic situations like this. Since LoadLeveler cannot magically know how much disk, memory, etc, your job will need, you have to tell it.

Each job needs an input file (as usual) and a special LoadLeveler File (I call them .cmd). Examples of .cmd files on cgate in ~sherrill/llscripts or just click here.

Submission directions:

• ssh cgate
• cd to directory in chem
• llsubmit xxx.cmd
• check status by llq -u [username]

To check jobs graphically:

• ssh cgate
• xloadl &
• On the linux boxes, a shortcut to the above two steps is sp2q

Job classes: When submitting jobs, use the appropriate class. This helps the machine keep running at maximum efficiency.

Note: You are strongly encouraged to begin using the ConsumableMemory option to help the SP2 avoid scheduling more jobs than will fit on a given node. We have currently set up the common nodes to accept "spillover" jobs from classes sio and scpu, but the common nodes only have 1GB (1000 MB) of memory. You can specify 500MB of memory by adding ConsumableMemory(500) to the # @ resources = ... line of your LoadLeveler file (*.cmd file).

Note: Always specify the following default values: Scratch(#), ConsumableCpus(#), and ConsumableMemory(#). Loadleveler only recognizes default values for particular classes and combinations of options.

    #!/bin/[shell]

    # @ job_type       =  serial
                          parallel

    # @ input          =  /dev/null

    # @ output         =  *.stdout

    # @ error          =  *.err

    # @ initialdir     =  path

    # @ notify_user    =  email address

    # @ startdate      =  mm/dd/yyyy
                          hh:mm:ss

    # @ class          =  scpu
                          sram
                          sio
                          squick
                          sinter
                          slong

    # @ notification   =  complete
                          always
                          error
                          never
                          start

    # @ checkpoint     =  no
                          user
                          system

    # @ restart        =  no
                          yes
 
    # @ requirements   =  (Arch == "power3")
                          (OpSys == "AIX43")
                          (Feature == "sa" )
                          (Feature == "sb" )
                          (Feature == "sherrill")
                          (Feature == "scratch2")

    # @ node_usage     =  not_shared
                          shared

    # @ resources      =  Scratch(#)           // # in pairs  default 0
                          ConsumableCpus(#)    //             default 1
                          ConsumableMemory(#)  // # in MB     default 256

    # @ queue          =  

    # @ node           =  #

    # @ tasks_per_node =  #

    [executable] [arguments]