Sampling
Sampling profilers works based on the logic that the part of a program that consumes most of the time (or that triggers the most occurrence of the sampling event) have a larger number of samples. This is because they have a higher probability of being executed while samples are being taken by the CPU Profiler.
Sampling Interval
The time between the collection of every two samples is the Sampling Interval. For example, in TBP, if the time interval is 1 millisecond, then roughly 1,000 TBP samples are being collected every second for each processor core.
The purpose of a sampling interval depends on the resource used as the sampling event:
- OS timer — the sampling interval is in milliseconds.
- PMC events — the sampling interval is the number of occurrences of that sampling event.
- IBS — the number of processed instructions after which it will be tagged.
Smaller sampling interval increases the number of samples collected and the data collection overhead. Since, the profile data is collected on the same system in which the workload is running, more frequent sampling increases the intrusiveness of profiling. A very small sampling interval also can cause system instability.
Sampling-point: When a sampling-point occurs upon the expiry of the sampling-interval for a sampling-event, various profile data, such as Instruction Pointer, Process Id, Thread Id, and Call-stack will be collected by the interrupt handler.
Event-Counter Multiplexing
If the number of the monitored PMC events is less than or equal to the number of available performance counters, then each event can be assigned to a counter and monitored 100% of the time. In a single-profile measurement, if the number of monitored events is larger than the number of available counters, the CPU Profiler time-shares the available HW PMC counters. This is called event counter multiplexing. It helps monitor more events and decreases the actual number of samples for each event and thus, reduces the data accuracy. The CPU Profiler auto-scales the sample counts to compensate for this event counter multiplexing. For example, if an event is monitored 50% of the time, the CPU Profiler scales the number of event samples by factor of 2.