See Also
The performance problems in your program could happen during a particular time period during the program's execution. Most of the Analyzer displays aggregate data for all times. The Timeline tab presents a graphical display of the events recorded in your experiments with time on the horizontal axis, so you can visually locate time periods that are of interest.
You could use the Timeline in many ways to find problems. Here are a few:
-
Scan the Samples bar for a pattern that represents time spent in a state
other than User CPU. Filter the data to display information only for the
interesting samples (in the standalone Analyzer, choose View
Filter Data; in the IDE, choose Analyze
View
Filter Data). Examine timing metrics other than User CPU
(in the standalone Analyzer, choose View
Set Data
Presentation; in the IDE, choose Analyze
View
Set Data
Presentation) in the Functions tab to find the functions responsible.
-
For a multithreaded application, scan the Synchronization Data bars to locate time periods where LWPs are spending significant amounts of time waiting.
-
For an MPI application, compare MPI Data bars between experiments to locate time periods where interprocessor communication could be a bottleneck.
-
Scan the Clock Data bars for each LWP in a multithreaded application to locate time periods where LWPs are idle.
Can't find what you are looking for?
Submit your comments at http://www.sun.com/hwdocs/feedback.
Legal Notices