Issue: When planning and running large workflows, there are some scalability issues to be aware of. During the planning stage, Pegasus traverses the graphs multiple times, and some of the graph transforms can be slow depending on if the graph is large in the number of tasks, the number of files, or the number of dependencies. Once planned, large workflows can also see scalability limits when interacting with the operating system. A common problem is the number of files in a single directory, such as thousands or millons input or output files.
Solution: The most common solution to these problems is to use hierarchical workflows, which works really well if your workflow can be logically partitioned into smaller workflows. A hierarchical workflow still runs like a single workflow, with the difference being that some jobs in the workflow are actually sub-workflows.
For workflows with a large number of files, you can control the number of files in a single directory by reorganizing the files into a deep directory structure.