To get the full features of POMS, when creating a CampaignLayer, you should have the following information and provide it to POMS, either in the CampaignLayerDefinition or in the CampaignLayer:
- Name for the general type of CampaignLayer for the definition, and for this specific instance (i.e. MyExperimentMonteCarlo and MyExperimentMonteCarlo_v4_3_purple_events)
- Job launch details
- Software launch script that you use
- what VO Role to use (i.e. Analysis, Production, etc.)
- Account/host you wan the launches run from
- parameters to software launch script to set
- SAM dataset
- software version, that will end up in SAM metadata as application version
- Dataset(s) describing all files this CampaignLayer will process
- How you want the dataset sliced into pieces for submissions; currently supported options:
- new -- files added since last launch
- mod_n -- (where n is an integer) sliced into n chunks.
- draining -- don't slice it, the dataset is a draining dataset which already excludes already processed files
- list of types of files the CampaignLayer generates, (i.e. if your program generates a processed data file and a histogram, how do we tell them apart?)
- specify how many recoveries you want run, and what sort of recovery dataset you want to build on each recovery
- if this CampaignLayer depends on the output of another campaignLayer, which one, and what file patterns.
- if the jobs have a specific number of input/output files (i.e. it reads only one file and writes only one) what those numbers are.
In future, we may also want to know
- memory/disk requirements
- expected job runtime
and we may have the option to increase memory usage requested for recovery jobs.