I want to make a point on FME handling of memory ressources and why I think it is not optimum. I am referring to Windows environment (as I have no experience with FME on other OS)
In the frist part of any FME log file some remarks can be found about how much physical and virtual memory it available and what limits will be used by FME for start/stop of requiring/freeing memory.
As described here(https://community.safe.com/s/article/Performance-Tuning-FME-FME-Memory-Management) FME will decide ressource usage depending on virtual memory available.
I think this is counter-productive in situations when more than one large FME process will fight for the same limited amount of memory.
In practise I see windows machines with 64Gb physical memory and 264Gb virtual memory quite often. In such an environment we run 3-4 very large FME processes in parallel and each one thinks 264Gb is available, which is not true of course.
We run into serious performance issues in this way, with physical memory completely used (99%) and hanging fme processes, as this relies on OS swapping, which is very bad on Windows.
This does not occur, if the virtual memory is reconfigured to be only a little larger than physical memory. So we have to advice our power customers to reconfigure their Windows System for small extra virtual memory only because FME is following a "strange" memory policy.
This is an undesirable situation, which cannot be followed in every customer situation as other applications may need the large extra virtual memory configuration.
So I want to really question the fact, why FME is optimizing memory usage based on virtual memory and not on physical memory.
Maybe a suggestion could be to influence the memroy policy to stay closer to the physical than the virtual limit by some parameter/environment variable or whatever.
Repeating myself: What I want to avoid is to tell our customers that they need to reconfigure their Windows OS setting only due to FME reasons.
