Event-based Sampling (EBS)Event-based Sampling (EBS)Use event-based sampling (EBS) to identify system-wide software performance problems caused by processor events, such as Cache Misses and Mispredicted Branches. From the EBS data, you can determine which process, thread, module, function, and source line in your program generated the most processor events, and if any of those events impacted the performance of your program.
While configuring the sampling collector for event-based sampling, the VTune(TM) Performance Analyzer determines the type of processor in your system and displays the corresponding events supported by your processor. Note that when configuring the sampling collector from the command line, you must type vtl -help -c sampling to view the corresponding events supported by your processor.
You can specify an event group containing a set of related events, or specify a single event and collect samples based on your selection. Performance analysts at Intel have identified and grouped key events that serve as a good starting point for performance tuning.
The VTune analyzer provides several predefined event ratios recommended for use by performance analysts at Intel. If two or more of the events you select form a useful event ratio, and a formula for this ratio has already been predefined, then the VTune analyzer displays the additional data for the event ratio in your sampling view. This data is displayed even if you did not actually configure the sampling collector to use the predefined event ratio. In order to get pertinent information on the performance of your application, use the predefined event ratios while configuring for event-based sampling. See Selecting Event Ratios for instructions on how to include event ratios when configuring the sampling collector for event-based sampling.
Event ratios can provide insightful data on the changes in the performance of your application. For example, the number of Branch Mispredictions Retired over the total number of Instructions Retired can provide more relevant data than each of the individual events.
