"So, from your explanation, what I conclude is that we can't increase the CPU utilisation precisely i.e. 50% then 100%, is that correct?"
No, I'm not saying that.
I'm saying it's almost impossible to accomplish that, without, perhaps, using traffic volume management, because we, well I, don't KNOW how to, or even if it's possible, as a user of the switch.
Further, if you do use the traffic volume approach, by design, (ideally) switch traffic is forwarded by ASICs, not the CPU (there's exceptions to that too, but do we KNOW exactly what they are?).
I'm also saying, for lack of KNOWLEDGE, we, again, at least I, don't KNOW all the interrelationships of the switch's components, so we (I) don't KNOW what to expect (like when something is going to "break") such as in your SNMP case.
Lastly, I'm saying, if you can overcome all the forging (which might be possible), you might not obtain any truly worthwhile information for your stated goal of temperature analysis.
Ponder, on that last point, how often do you see a "professional" network device review analysis temperatures vs. CPU utilization?
Back to my car analogy - you might see some report, that Brand X car, at 30 MPH takes 400' to stop, but at 60 MPH takes 1200' to stop, but do you see a Brand X car report describing, at 30 MPH engine shows a coolant temperature of 200 degrees but at 60 MPH engine shows a coolant temperature of 201 degrees? (Another point I've been trying to make, might a car engine run hotter, even much hotter, at 60 MPH vs. 30 MPH? Yup, very likely it does. But the whole point of the car's cooling system is to carry away the excess heat. If you do see a major rise in coolant temperature, you've got a coolant system problem. Likewise for your switch, its cooling system is designed to avoid the interior of the switch baking in its own heat. I.e., by its design, you might not see any huge temperature increase between CPU running at 1% vs. 100%!)
To recap, trying to PRECISELY control your switch's CPU utilization, is likely very difficult to nearly impossible, and from your described goals, even if you could manipulate your switch's CPU as desired, with just on-board temperature sensors, I doubt (but, I don't KNOW) you'll would get any useful information.
To put it another way, IMO, for your goals and equipment, you're wasting your time. However, I could be wrong, and even if I'm not, it's YOUR time, not mine.
If you do succeed, please post what you did and how you did it; it may be interesting to read.
