The Java language removes several problem areas for developers, compared to C++ development, by its elimination of memory allocation and pointer management. While this generally makes Java programs more stable than their C++ brethren, it can often lead to the misconception that Java programs don't need to be optimized or profiled. Nothing could be further from the truth, especially when you consider that Java is being used to build enterprise-class applications, which require high performance. KL Group has built a strong reputation for supplying Java components to many of the leading software vendors, and they have extended their Java presence with JProbe Profiler.
KL Group delivers the software directly from their Web site for download, as well as offers the software on CD-ROM. JProbe is available for Windows 95/NT and installs with an InstallShield executable. I was able to get the software installed in a matter of minutes, and the complete installation required only about 14 MB of disk space. JProbe requires the JDK 1.1.5 version to run, but you can profile any application that was written for either the JDK 1.0 or 1.1 series virtual machine. If you have already installed the JDK 1.1.5 virtual machine, you'll still need to install the version that comes with JProbe, since it's a specialized version of the standard 1.1.5 JDK virtual machine. The runtime environment for JProbe allows you to set CLASSPATH definitions for individual programs as needed, so it's not necessary to modify your Java development environment to get started.
Uncovering Performance Bottlenecks
Java as both a platform and a language offers a number of advantages in terms of portability and standardization. Programmers across the globe are migrating to Java from a variety of other languages and platforms. For many of us the move to Java may also be our first real, hard-core experience with object-oriented programming.
As a result, the programs you write may have hidden performance bottlenecks that slow down processing and eat memory. KL Group's JProbe is designed to uncover these hidden performance anomalies including such common problems as excessive object creation, method calling and thread creation, and inefficient memory usage. The Profiler collects timing information and memory data as you run your Java programs, and JProbe supports most of the leading Java development environments. I used Oracle's JDeveloper product to create a few sample Java programs for testing. Once you've compiled your code in your favorite development tool, start the Profiler and select the "Run" menu. I was quickly able to search for possible performance problems in my code by using the "Memory Usage Monitor," as shown in Figure 1.
The memory monitor charts memory as it's used by the program and is generally the first place you start when you use the Profiler. KL Group includes a number of sample program runs with JProbe, and I'd advise you to start by using these examples with the Profiler's online help to get an overview of the Profiler's capabilities. You have a choice of running a program through to conclusion as one long run, or you can divide the performance data into specific chunks by using "snapshots." I found the snapshots invaluable as a tool for comparing the first pass of a program with subsequent executions through the same set of code. They were especially helpful tracking down problems with event-driven code. I was able to use the memory monitor to drill down into details of my code quickly and easily. The snapshots can be used to track calling relationships, as shown in Figure 2.
You can see in the diagram that the profiler can show a hierarchical display of method calls, which can help you expose the most expensive methods. If you need to, you can click down to the source code directly from the diagram, and the graphical interface uses colors to highlight the more expensive method calls.
If you choose to view the source within the Profiler, you can see how much resource is used by each routine in a panel next to the source window. The Profiler makes it easy to shift between multiple snapshots, and you can save the entire set of snapshots and program definitions for reuse at a later time. I was impressed with how quickly I was able to find problems in my own programs, but I'll take a pass on telling you just how bad my coding actually was before JProbe got ahold of it!
Performance and Usability
KL Group claims to have improved the performance of JProbe Profiler by a factor of 10 with this release. I didn't test any large or long-running programs with the Profiler, but JProbe ran briskly enough on my development platform. I couldn't see any easy way to test programs out in batch, although I was able to save test sessions for later analysis, and you can save programs for reexecution as well. The online documentation with JProbe Profiler is adequate for using the product, but it's a little weak in the area of interpretation. The help files have a tendency to expect that many of the performance numbers will be self-explanatory, and novice programmers may find that it'll take some practice to get a handle on interpreting results. I'd encourage you to make use of the tutorial and sample files before tackling any of your own code.
There are several products on the market that purport to provide performance profiling for Java. However, when I quickly searched the various Java newsgroups for user opinions, I found that many programmers had good things to say about JProbe Profiler. One programmer in particular claimed that JProbe had helped him to get a hundred-fold improvement in performance. My informal Web survey seemed to indicate that JProbe is clearly the leader in Java profilers, and from my brief experience with the product, I'd be inclined to agree.
About the Author
Jim Milbery is an independent software consultant based in Easton, Pennsylvania. He has over 15 years of experience in application development and relational databases. Jim can be reached at [email protected] or via his Web site at www.milbery.com.