Many times, I have a workflow that reads multiple tables from one database, performs a few simple tasks to each one individually, and writes to another database. If there are 10 tables, it completes one before moving on to the next. This can add up to long processing times in the end. However, if I split that job up into 10 separate workbenches and run them concurrently on FME Server, the total processing is drastically reduced since they are run in parallel. Unfortunately, that method creates a data management nightmare.
My suggestion is to allow parallel reads and writes within one workbench (when those reads/writes don't depend on the other reads/writes in the same workbench).
Obviously this wouldn't apply if there are table joins or any transformers that hold features.
Example: The screenshot below is a job that takes nearly 24 hours to run due to large amounts of data. However, no single table takes more than a few hours. But since FME runs in "serial" mode, all those hours are added together in the end.
