HP UX Java Software Percent (%) CPU Utilization

Figure 8-17 Monitoring Metric: Loaded Classes

Percent (%) CPU Utilization

Displays total system and process CPU consumption.

Guidelines

•Percentages are displayed as number of CPUs x 100%.

•Excessive use of CPU resources (greater than 80% of the total number of CPUs) may indicate that the application load limit is close, even though the application may appear to be performing well. At higher consumption rates, CPU consumption can become a bottleneck to good performance.

•When system CPU consumption is significantly higher than process consumption, this may indicate that “alien” or undesired processes are using CPU resources that the preferred application could be using. It may also indicate that the application is abusing the operating system kernel.

136 Using Visualizer Functions

Contents

Page Page Table of Contents About This Document 1 Introducing HPjmeter 2 Completing Installation of HPjmeter 3 Getting Started 4 Monitoring Applications 5 Profiling Applications 6 Analyzing Garbage Collection Data 7 Using the Console 8 Using Visualizer Functions Page Page 9 Understanding How HPjmeter Works 10 Troubleshooting A Quick References Index About This Document Intended Audience Typographic Conventions Additional HPjmeter Documents Related Information Publishing History HP Encourages Your Comments 1 Introducing HPjmeter Features Concepts JVM Agent Node Agent 2 Completing Installation of HPjmeter Platform Support and System Requirements Agent Requirements Console Requirements Completing the installation File Locations Attaching to the JVM Agent of a Running Application Configuring your Application to Use HPjmeter Command Line Options Preparing to run Java Example Usage JVM Agent Options bytecode instrumentation JVM agent filters Console filters •You can change the default behavior by using the include option. For example: -agentlib:jmeter=include=com.ibm.ws,exclude=com.ibm.ws.io -agentlib:jmeter=exclude=sun.jdbc.odbc:other-package-names sun.jdbc.odbc not Page Security Awareness Securing Communication Between the HPjmeter Node Agent and the Console Securing Communication Between the JVM and the HPjmeter Node Agent 3 Getting Started Are You Monitoring an Application or Analyzing Collected Data Using HPjmeter to Monitor Applications Configure and Start Your Application Confirm that the Node Agent is Running 1.Choose Connect from the File menu or click Connect to server [ ] Optionally Optional Port 4.Click Connect •— A JVM that has already loaded an HPjmeter JVM agent 1.Choose Connect from the File Menu or click Connect to server [ ] Connect to Server 4.Check the box labeled Use SSH Options The Connect to Server dialog expands to reveal required SSH attributes 5.Type the IP address External IP) of the SSH delegate Set Session Preferences When the Optional Port Box Is Not Used Interdependencies Exist Among Some Metrics and Between Some Metrics and Alerts: Once all selections are made, click Changing Session Preferences During a Session View Monitoring Metrics During Your Open Session Using HPjmeter to Analyze Profiling Data with zero preparation and no interruption of your application 1.Configure your application Systems 2.Run the application to create a data file 3.Start the console from a local installation on your client workstation Using HPjmeter to Analyze Garbage Collection Data Monitoring Demonstration Instructions Memory Leak Applications Thread Deadlock Sample The demo application is available from the HPjmeter installation directory: Source: $JMETER_HOME/demo/DL1.java Binary: $JMETER_HOME/demo/DL1.jar Use —jarDL1.jar with the run_simple_jvmagent script to start the sample Figure 3-2Thread Deadlock Example Display 4 Monitoring Applications Controlling Data Collection and Display Setting Data Collection Preferences Managing Node Agents Managing Node Agents On HP-UX Node Agent Access Restrictions Running Multiple Node Agents Saving Monitoring Metrics Information Saving Data from the Console Naming Monitoring Data Files Diagnosing Errors When Monitoring Running Applications Identifying Unexpected CPU Usage by Method Viewing the Application Load Checking for Long Garbage Collection Pauses Checking for Application Paging Problems Identifying Excessive Calls to System.gc() Reviewing the Percentage of Time Spent in Garbage Collection Page Checking for Proper Heap Sizing Confirming Java Memory Leaks Determining the Severity of a Memory Leak Identifying Excessive Object Allocation Identifying the Site of Excessive Object Allocation Identifying Abnormal Thread Termination Identifying Multiple Short-livedThreads Identifying Excessive Lock Contention Identifying Deadlocked Threads Identifying Excessive Thread Creation Identifying Excessive Method Compilation Identifying Too Many Classes Loaded Using the JMX Viewer Understanding the JMX Summary View Figure 4-5Appearance of the Summary JMX Viewer When First Opened JMX Memory Tab The following image demonstrates some of the features available on this tab Region Details Enter Perform GC JMX Threads Tab Threads Changing Mbean Values and Monitoring the Result Using the Functions in the JMX Server View Use Filter The MBean Attribute Tab Attributes Detail Viewer Editable value monitoring metric The MBean Operations Tab Operations Operation_Name getThreadCpuTime The MBean Notifications Tab The MBean Information Tab Information Page 5 Profiling Applications Profiling Overview Tracing Sampling Tuning Performance Preparing a Benchmark Collecting Profile Data Profiling with -Xeprof Profiling with Zero Preparation Profiling with -agentlib:hprof -Xrunhprof Table 5-2Supported -agentlib:hprof options Naming Profile Data Files –Xeprofand –agentlib:hprofProfiling Options and Their Corresponding Metrics Page •Profiling with -Xeprof(page 61) •Profiling with -agentlib:hprof(page 63) Approaches to Analyzing Performance Data Looking at the Data from the Bottom Up Looking at the Data from the Top Down Looking for Inefficiencies in Memory Usage Considerations in Interpreting the Data Locating Summary Information for Saved Data Sets Adjusting Scope Comparing Profiling Data Files Scaling Comparison Data Reading Profiling Histograms Key to Thread States Reported by Interpreting the Histogram Presentation Using Call Graph Trees Interpreting Call Graph Data Example of Node Color Display Options for Manipulating the Call Tree Display Using Heuristics to Locate Possible Hot Spots Page 6 Analyzing Garbage Collection Data Obtaining Garbage Collection Data Data Collection with -Xverbosegc Table 6-1Supported -Xverbosegc options for Java 1.5.0.04 (continued) java ... -Xverbosegc:help Table 6-2Fields Captured in GC Log Data When Using -Xverbosegc print options Page Collecting GC Data with Zero Preparation Data Collection with -Xloggc Naming GC Data Files -Xverbosegcand -XloggcOptions and Their Corresponding Metrics Comparing Garbage Collection Data Files Basic Garbage Collection Concepts Key to Garbage Collection Types Recognized by HPjmeter Understanding the Summary Presentation of GC Data Figure 6-2Summary Panel Showing Garbage Collection Statistics Heap Capacity Peak Usage of Capacity GC Activity Summary Duration of the Measurement Understanding the System Details Captured with GC Data Page Page 7 Using the Console Starting the Console Starting the Console On HP-UX Starting the Console On Linux Starting the Console On Microsoft Windows Data Representation Icons and Their Meaning Table 7-2JVM Agent Connection Icons (continued) Open and Cached Sessions Cached Application Data Table 7-3Session State Icons Time Slice Entries Console Tool Bar Buttons Console Menu Choices ◦Remember Main Window Location ◦Cascade Metric Windows ◦Show Console Guide ◦Confirm Save When Closing the Main Window —Standard Preferences Console Guide Status Bar Setting Monitoring Session Preferences Specifying Metrics to Collect for Monitoring Enabling Monitoring Alerts Specifying Filters for Monitoring Viewing Monitoring Data in HPjmeter Using Alerts Page Using the Alert Controller enable the alert in the session preferences Figure 7-9Alert Controller Window You can use the Alert Controller in these ways: Deactivate or reactivate enabled alerts log file Editing E-mailNotification Attributes Responding to Alerts Figure 7-12Abnormal Thread Termination Alert Visualizer •Poor application response •The entire application terminates Uncaught Exception Statistics These statistics are derived from more than one metric Show Packages Show Stacktraces Excessive Compilation Alert Expected Out Of Memory Error Alert GC Duration Notification Heap Usage Notification Java Collection Leak Locations Alert Array Leak Locations Alert Process CPU Usage Alert 5.Click Apply to activate checking for this threshold When the set threshold is met, an alert notice is posted on the console Percent CPU Utilization For more information, see also: •Set Session Preferences (page 26) Thread Deadlock Alert Unfinalized Queue Growth Unfinalized Queue Growth 8 Using Visualizer Functions Visualizer Behavior When Monitoring Behavior or Analyzing Data Locating Information About a JVM and its Environment Using Monitoring Displays Monitor Code and/or CPU Activity Menu Page •The application runs on a single CPU system •The application does not consume a lot of CPU •The application consumes CPU exclusively in non-profiledmethods •Data Sampling Considerations (page 188) •Time Units (page 69) Page Monitor Memory and/or Heap Activity Menu Garbage Collections young generation GC Duration Percentage of Time Spent in Garbage Collection Unfinalized Objects Allocated Object Statistics by Class Figure 8-8Monitoring Metric: Allocated Object Statistics by Class •“Data Sampling Considerations” Allocating Method Statistics Shows the methods that allocate the most objects Current Live Heap Objects Refresh Live Objects Figure 8-10Monitoring Metric: Current Live Heap Objects Table 8-1Data Shown in Current Live Heap Objects Visualizer To select a portion of the data Copy Selection to Buffer Monitor Threads and/or Locks Menu Lock Contention already acquired by another thread Running All remaining cases •Lock Contention appears as red in the display Lock Contention Provides lock contention statistics The reported data is cumulative over the lifetime of the session Figure 8-14Monitoring Metric: Lock Contention Monitor JVM and/or System Activity Menu Method Compilation Frequency Loaded Classes Percent (%) CPU Utilization Viewing Profiling or GC Data in HPjmeter Using Profile Displays Page Page Menu Choices Profile Code and/or CPU Activity Exclusive Method Times (CPU) exclusive times Exclusive Method Clock Times Call Graph Tree with Call Count Call Graph Tree with CPU Call Graph Tree with Clock Time Inclusive Method CPU Times Inclusive Method Clock Times Average Exclusive Method CPU Times Average Exclusive Method Clock Times Average Inclusive Method CPU Times Average Inclusive Method Clock Times Starvation by Method These are possible reasons for reported starvation: Profile Memory and/or Heap Activity These metrics – Objects Created by Method Created Objects (Count) Created Objects (Bytes) •Allocation Site Statistics (page 161) finalize() •Threads Histogram (page 152) Reference Graph Tree Shows all live objects and references between them Expanding the nodes shows which references a given object holds Figure 8-19Reference Sub-Treesby Size Visualizer – Attributes table Object and Primitive Value Tables The Attributes table shows the names and values of the Page Page Class Loaders Figure 8-24Class Types Pop-upWindow Use Mark Find •Reference Graph Tree (page 146) Profile by Threads Profile by Locks Contested Lock Claims by Method All Lock Claims by Method Lock Delay - Method Exclusive Lock Delay - Call Graph Tree Lock Delay - Method Inclusive Lock Contention Ratio by Method Average Exclusive Method Lock Delay Exclusive Method Lock Delay / Clock Time Average Inclusive Method Lock Delay Using Heuristic Metrics from the Estimate Menu Using Specialized Garbage Collection Displays Heap Usage After GC Duration (Stop the World) Cumulative Allocation Creation Rate Allocation Site Statistics combined separated Combined Sets Show Clock Time Figure 8-31GC Metric: Combined Sets of Allocation Site Statistics Filter Page Filter Field Descriptions: Time All % Allocated Bytes Allocated How a Snapshot of Allocation Sites Statistics is Shown in GC Visualizers User-Defined X-YAxes Multiple User-Defined Figure 8-38Multiple User-DefinedMetrics Figure 8-39Multiple User-DefinedMetric Configuration Box Metric Selection Mode Single Multiple Glance Data Page Figure 8-41Glance Data Metric Configuration Box tab describes the metrics available for display in the metric configuration pop-up box of the Table 8-2Metrics Shown in Glance Adviser Visualizer (continued) Glance System Call Data Page Using Visualizer Tool Bars Common Tool Bar Buttons Tool Bar Buttons for Manipulating Tabular Data Tool Bar Buttons for Manipulating Graphical Data Special Button or Other Gadget Functions Mark Mark to Find Marked java.util.BitSet.recalculatedUnitsinUse Using the Marked Object List on the Console To use this feature, do the following: •Java package name •Java class name •Java type name (classes, primitive types, and arrays) •Java method name •Java object reference (ID) Clear Find All Pause or Resume Graphical Time-basedScrolling Changing Time Interval in GC Data Visualizers Page Page Changing Time Interval in Monitoring Visualizers Change Color Selection for Histogram Display Page 9 Understanding How HPjmeter Works Performance Overhead on Running Applications Application Server Startup Time Monitoring Overhead Profiling Overhead and Intrusion Data Sampling Considerations Using Confidence Interval to Indicate Sample Validity How Memory Leak Detection Works Tapping in to Standard Management of the Java Virtual Machine Page 10 Troubleshooting Documentation and Support Identifying Version Numbers Installation Console JVM agent Node agent Zero Preparation Profiling Unexpected Behavior in Visualizers A Quick References Connecting to the HPjmeter Node Agent Determining Which HPjmeter Features Are Available With a Specific JVM Version Default Location of Xverbosegc and Allocation Site Statistics Data Files Glossary Page Page Index