This chapter is from the book
This chapter is from the book
This chapter is from the book
High CPU Utilization
Sometimes an application simply cannot meet service level performance or scalability agreements even though performance efforts have reduced system CPU utilization, have addressed lock contention, and other optimization opportunities have been addressed. In such cases, doing an analysis of the program logic and the algorithms used is the direction to take. Method profilers such as the Performance Analyzer or NetBeans Profilers do a good job at collecting information about where in general an application spends most of its time.
The Performance Analyzer's Call Tree tab is good at providing an application's hottest use case by showing the call stack trees. This information can be leveraged to answer questions in a more abstract way, such as how long does it take the application to perform a unit of work, or perform a transaction, use case, and so on so long as the person looking at the profile has sufficient understanding of the implementation to be able to map a method entry point as the beginning of a unit of work, beginning of a transaction, use case, and so on. Being able to analyze the profile in this way provides the opportunity to step back, look at a higher level, and ask questions such as whether the algorithms and data structures being used are the most optimal or are there any alternative algorithms or data structures that might yield better performance or scalability. Often the tendency when analyzing profiles is to focus primarily on the methods that consume the most time in an exclusive metric kind of way, that is, focusing only on the contents of a method rather than at a higher level unit of work, transaction, use case, and so on.