every time the overload reaches 600%++ (2-3min) I get a bad_failure on some tags (not all) I find it strange that it is only on some of them that's why I have this question... why only on those? . also have you checked that there are no scripts associated with those tags.
Look at your mean response time. That is typical when a true serial modbus link is behind your target device and is running at 9600 baud or 19200. You simply cannot read that much data at the desired pace.
He's saying it's Modbus TCP (At least hopefully that's the case and that the device named MRTU means Modbus Roof Top Unit since it's HVAC and not Modbus RTU). Either way, it's definitely something that needs more optimization. I just had to do some similar cleanup on an Allen Bradley project where everything was being polled all the time when most of the data wasn't used except on faceplate. Once I changed those over to leased tags with 0 for the normal "rate", it stopped polling tags it really didn't need and went from 400-500% overloaded to 0-1%.
it might make some sense with what I am trying to solve because most of the tags are associated to a tag group (with the configuration I will show in the next image) where clearly they will not be used unless the user interacts with them, while they will be on standby which I think is the point you are referring to with no data wasn't used except on faceplate.
in this case, what you are proposing is to switch the Direct mode to Leased?