The key to properly sizing a machine (or machines) where the ML Server(s) is/are running is based on three things in general :

1. The average and/or maximum number of concurrent syncrhonizing you expect to be active at any given time.
2. The average and/or maximum size of the upload and/or download
3. Most importatnly, YOUR testing in YOUR environment, because every situation is different.

In a perfect world you would love to have enough physical memory available to the MobiLink Server(s) to hold all the data for all active synchronizations. You need to decide how paranoid you want to be, and whether you are sizing so that you can handle the largest downloads with near maximum concurrent synchronizations, or whether you are sizing based on averages.

When considering on premise versus SAP Cloud Platform, the most important thing to consider is that disk space on premise is super cheap, but disk space in any hosted environment is not. When MobiLink runs out of physical memory to store upload/download data, it will spill the extra data to disk, so having enough physical memory and disk swap space to handle the worst case scenario is important. When considering the worst case scenario, you should also have in the back of your mind that user or two (or seven) that hasn't synchronized in a long time, and may have a larger thatn normal upload or download.

Specifics on sizing/pricing of resources in the SAP Cloud Platform or reference customers that have documented and are willing to share their experiences is something you'll likely need to talk to sales about. This forum is mostly filled with technical people.

Reg