We have a monitoring package that offers two variations for monitoring a Windows server: local agent and WMI. By local agent, I mean that there is a service running on the server, collecting data and sending it to a manager. The WMI approach means one data collector can use WMI to get data for many servers. Certainly it is easier to upgrade the WMI approach, but the local agent is able to monitor even if a network connection is down. So there are some tradeoffs.
I want to ask specifically about the accuracy of reported performance data. Are local agents (e.g. perfmon?) more accurate than the WMI approach? Can the accuracy of WMI be improved sufficiently by increasing the frequency of "polling"?