Starting Pegasus 4.0 , all compute jobs ( single or clustered jobs) that are executed in a non shared filesystem setup, are executed using lightweight job wrapper called PegasusLite.
When PegasusLite starts on a remote worker node to run a compute job , it performs the following actions:
Discovers the best run-time directory based on space requirements and create the directory on the local filesystem of the worker node to execute the job.
Prepare the node for executing the unit of work. This involves discovering whether the pegasus worker tools are already installed on the node or need to be brought in.
Use pegasus-transfer to stage in the input data to the runtime directory (created in step 1) on the remote worker node.
Launch the compute job.
Use pegasus-transfer to stage out the output data to the data coordination site.
Remove the directory created in Step 1.