Loupe supports two distinct models for recording metrics:
Use to record data about individual events without any effort to do summarization or aggregation. This is often simpler than using sampled metrics which require the data be aggregated within your application, and can produce a wider range of analysis options. With an event metric, you record the characteristics of each individual occurrence you want to track. For each of the numeric values, you can specify how to aggregate multiple events together for analysis. For qualitative (non-numeric) values, the count of events will be used.
Web server traffic logs represent a good example of an Event metric. With them, a single record is recorded each time the web server processes a request including a number of columns of information. From this, a wide range of analysis can be performed, and the decision on what analysis are useful can be deferred outside of the development effort.
Used to record a single, summarized value on a periodic basis. Whenever a sample is recorded, the value is presumed to remain constant until the next sample is recorded. This works best when the data for each individual event is either unavailable or would be to inefficient to collect. To do this, the application has to do some degree of calculation or aggregation on its own and the useful trends have to be determined during development.
Windows Performance Counters are collected as Sampled metrics. Processor Utilization represents a good example of a sampled metric: It would be infeasible to record each time the processor was used for a task, and the underlying data isn't available.