Configure the amount of wall-clock time the Intel® VTune™ Profiler waits before collecting each sample (sampling interval).
The sampling interval is used to calculate the target number of samples and the Sample After value (SAV). Increasing the sampling interval may be useful for profiles with long durations or profiles that create large results. Typically, the size of the collected result is affected with such factors as duration, thread and core counts, selected analysis type, additional collection knobs, and application behavior.
You may change the default sampling interval as follows:
- Click the (standalone GUI)/ (Visual Studio* IDE) Configure Analysis button on the VTune Profiler toolbar.
- Select a predefined analysis type from the HOW pane or create a custom analysis type.
- Use the CPU sampling interval, ms field to specify the required interval.
For user-mode sampling and tracing types, specify a number (in milliseconds) between 1 and 1000. Default: 10ms. For hardware event-based sampling types, specify a number between 0.01 and 1000. Default: 1ms.NOTE:
For hardware event-based sampling types, the sampling interval serves as a simple SAV multiplier so that the default interval value of 1ms just leaves the SAV intact. The sampling interval value of 0.1ms divides the SAV for all events by 10 making them overflow 10 times more frequently. The sampling interval value of 10ms multiplies the SAV for all events by 10 making them overflow 10 times less frequently.
To determine an appropriate sampling interval, consider the duration of the collection, the speed of your processors, and the amount of software activity. For instance, if the duration of sampling time is more than 10 minutes, consider increasing the sampling interval to 50 milliseconds. This reduces the number of interrupts and the number of samples collected and written to disk. The smaller the sampling interval, the larger the number of samples collected and written to disk.
The minimal value of the sampling interval for the user-mode sampling and tracing collection depends on the system:
10 milliseconds for Windows* systems with a single CPU
15 milliseconds for Windows* systems with multi-core CPUs
10 milliseconds for Linux* 2.4 kernels
1,2,4 milliseconds for new Linux >= 2.6 kernels depending on the vendor
For driverless Perf*-based data collection on the targets running under Xen Hypervisor, the VTune Profiler automatically sets the sampling interval to 0 to switch to the integrated Perf sampling interval. This configuration provides more precise performance statistics in the hypervisor environment.