(Click to open topic with navigation)
You must modify your Moab configuration to enable data staging. In addition to the configuration steps described below, you might also consider customizing the configuration (including the associated scripts) to meet your site's specific needs.
For advanced configuration steps and options, see Configuring Data Staging with Advanced Options.
To configure data staging
Verify that your firewall and network are correctly configured to allow the scripts to operate as designed.
> yum install python-paramiko python-mock
> ls -l /opt/moab/tools/data-staging
You can copy and modify the reference scripts and configuration files to meet your specific needs. See the README file packaged in the data-staging directory for information about modifying these files.
SUBMITFILTER /opt/moab/tools/data-staging/ds_filter
The data staging filter checks the msub argument syntax to verify that the arguments make sense and are consistent; attempts a dry run connection via SSH and the file transfer utility to ensure that keys exist for the user on the necessary systems; and attempts to determine the size of the data that will be transferred.
You can customize the script to meet your specific needs; the file contains detailed comments illustrating its default behavior to facilitate its modification. If you replace or modify the submit filter, it is your responsibility to ensure that the same functionality described in the paragraph above is present in your filter.
Note that this filter has the DEFAULT_TEMPLATE name which should match the name of the master data staging template in moab.cfg. For more information, see Configuring Data Staging with Advanced Options.
Example 24-1: Non-grid
RMCFG[torque] Type=pbs
PARCFG[torque] GMETRIC[DATASTAGINGBANDWIDTH_MBITS_PER_SEC]=58
Partition torque has a transfer rate of 58 megabits per second. Moab uses the rate when it estimates the time it will take to stage data in and determine when to schedule the job that will use the data.
Example 24-2: Grid
RMCFG[m1] type=Moab
PARCFG[m1] GMETRIC[DATASTAGINGBANDWIDTH_MBITS_PER_SEC]=100
Partition m1 has a transfer rate of 100 megabits per second. Moab uses the rate when it estimates the time it will take to stage data in and determine when to schedule the job that will use the data.
NODECFG[GLOBAL] GRES=bandwidth:10
Data staging jobs can use up to 10 units of bandwidth on the system. You can specify the number of units consumed by each data staging job when you configure the data staging job templates.
DataStaging requires "JOBMIGRATEPOLICY JUSTINTIME" to ensure the workflow job ids are not altered upon submission.
Related Topics