Performance Monitoring offers to main benefits: Being alerted when a performance counter exceeds a threshold and collecting performance counters in a database. As with all system health features, we recommend that you create multiple packages when necessary.
Package Optimization
If you only collect performance data in a database for some computers, but need alerts from all computers then you can create two separate packages: One for alerts, and one for database logging.
In most cases you will want to create a general performance package that will monitor counters such as CPU usage, memory utilization, disk queue length and network usage and then set this package to be global. Create additional packages with performance monitoring objects for additional server software (e.g. IIS, SQL, Exchange) and apply them as needed.
Suggestions - Impact on Server Performance
The impact that performance monitoring has on your servers depends on the number of counters you are monitoring and how often you are reading performance data. In most cases the impact on your servers' performance will be insignificant.
Polling Interval
A small interval (e.g. 1 second, 2 seconds) will result in very accurate data but will put more stress on the monitored system than a larger interval (e.g. 5 seconds, 10 seconds). If you need accurate data then a polling interval of 5 seconds is a good compromise, 10 seconds should be enough for most other cases.
Threshold Alerts
The values entered in this section depend on the counter being monitored, but you should ensure that the time interval is not too short. If it is, then you will probably receive a lot of false positives (especially for high CPU usage).
Database
We recommend that you always check the "Log Average" checkbox when the database logging interval is higher than the polling interval (which it should be).
Let's assume that you are monitoring the CPU usage on a server and polling the data every 5 seconds and you have the database interval set to 5 minutes (even this seemingly large interval means that approximately 8640 entries will be written every month for this counter, per server). EventSentry will store all recent counter data for the entire database interval (5 x 12 = 60 values) and then write the average every 5 minutes to the database. If you don't set the "Log Average" option, then the current counter value - every 5 minutes - will be written to the database. This would obviously not create a very accurate picture of this performance counter.
