Clock-Based Profiling Data

Related Topics:

Callers-Callees Function Metrics	Annotated Source and Disassembly Code
Synchronization Wait Tracing Data	Hardware Counter Overflow Profiling Data

Clock-based profiling shows how much time, exclusive and inclusive, your program spends in each function or load object, and helps to identify where program bottlenecks are occurring. For an explanation of exclusive and inclusive metrics, see Exclusive, Inclusive, and Attributed Metrics.

Clock-based profiling data is converted into the following timing metrics for each function in the program:

• User CPU time. LWP time spent running in user mode on the CPU.
• Wall time. LWP time spent in LWP 1. This is the "wall clock time"
• Total LWP time. Sum of all LWP times.
• System CPU time. LWP time spent running in kernel mode on the CPU or in a trap state.
• Wait CPU time. LWP time spent waiting for the CPU.
• Text page fault time. LWP time spent waiting for a text page.
• Data page fault time. LWP time spent waiting for a data page.
• Other wait time. LWP time spent waiting for a lock or for a kernel page, or time spent sleeping or stopped.

Except for wall clock time, all metrics are summed across LWPs. Wall clock time is meaningless for multiple-program multiple-data (MPMD) programs.

By default the Function List display shows exclusive and inclusive user CPU time in seconds, sorted on exclusive user CPU time. The defaults can be changed using a defaults file. See the er_print(1) man page for details.

You can choose to display metrics as a percentage of total program metric, select other metrics to display and specify a different sort order in the Function List Metrics Dialog Box. See Choosing the Metrics Displayed and the Sort Order for more information.

You can also select a subset of experiments, samples, threads and LWPs for which to examine metrics. See Selecting Experiments, Samples, Threads, and LWPs for Analysis.