Server key performance counter to monitor

The thresholds set will usually be based on the server in question, what it's being used for, etc. You can download the demo of NetCrunch 8, which has out-of-the-box monitoring packs for Windows Server, SQL, etc and you can pull the thresholds and alerts that are set there, but again, that'll just give you a baseline. If you have time, NetCrunch can also do baseline thresholding for you, in that it will monitor the performance of your server over time, and create baselines for standard performance, and then you can set alerts at a percentage over baseline.

Thank you for this insight, however I would like to have just the basic thresholds to defined:
Ex: for CPU, disk storage and RAM which type of critical limitation I can set up?

Well, a typical threshold is 90%, but like i said, it would vary on the usage of the device. It all depends on how you're using the server, though. A server that is typically running around 50% may be encountering a problem that's worth notifying you if it hit's 70%, but using a basic threshold you wouldn't know that something is amiss. However, in all cases, you're gonna want to be notified at the 90% level. If possible, you'd want to get an average over time/check. One spike over 90% may not be bad, but if the server is checked every 5 minutes, and 3 out of 5 show over 90%, then maybe you'd like a warning. Really, it all comes down to usage, and how well you know your servers ;)

@Michael: thank you for this insight.
Concerning the RAM and SQL server which type of Key performance indicators I can take as a reference?

Ok, thank you!

NP. Also, an average over 3 checks would probably be perfect. Depending on how many devices you have, the robustness of your monitoring platform, and the hardware it's running on the interval between may change. But 5 minutes should be more than enough.