Hi Marco

I'm assuming you're talking about the memory usage of SLDataMiner? SLDataMiner is a 32bit process, meaning it can use a maximum of 4GB RAM.
Practically, we can see it usually crashes around 3.5GB.

Looking at your system in particular, an average of 1GB looks normal. So if it reaches 2GB, you might say there's an issue.
In terms of restarting, 3GB would be my final warning. This should give you time to schedule in a DMA restart (depending on the root cause of these memory increases).

How to achieve this?
Within the alarm template of your Microsoft Platform elements, The VMSize of SLDataMiner should be defined according to your own preference (or according to the above advice).
If this isn't enough, then you can create a correlation rule based on this/these alarms (if needed, specify the element linked to the DMA). Configure the rule to send a mail to you and/or your team, to warn about the situation and a DM restart should be scheduled.

Note
The above assumes you have a Microsoft Platform element per agent and its parameter "Poll Task Manager" is set to enabled.

Note2
The above is unrelated to the DataMiner version.

Note3
The memory and/or CPU usage of SL* processes can vary on how you use your DM system. So it's always a good practice to define the alarm template based on the system itself (and not just rely on any default values provided).

Note4
If you experience SLDataMiner or any other SL* process to increase unusual in size, it's always good to involve techsupport@skyline.be