In just a few years the Java language and platform has become the technical approach of choice for building complex, distributed and Web-enabled applications across the enterprise. Thanks to its cross-platform runtime environment, object-oriented development model, and facilities for working with object request brokers and other code components, Java is well equipped for building such applications.
Java enables software developers to provide seamless communication and application access to the rapidly growing world of Internet computers and communications devices, from UNIX servers and PCs to cell phones and beyond. IDC Research estimates that the Java products market will grow by 85% annually through 2004.
Java applications - especially those designed to work with other applications and components (often written in other languages) in a distributed environment - have different development requirements than traditional applications. For example, Java's execution model virtually eliminates traditional memory errors, but can introduce performance problems stemming from poor resource utilization. For those used to addressing traditional programming errors and other issues, Java's unique execution model and language characteristics may make building error-free and efficient applications more difficult.
The distributed nature of many Java applications can also make it difficult to pinpoint performance issues or diagnose programming errors. As a result, Java developers need software development tools and techniques for viewing and analyzing code execution on multiple systems on the network and from multiple code bases.
Issues Surrounding the Development of Distributed Java
Java is unique - it's a mainstream programming language that works like no other. Its rules aren't well understood yet by many application developers. Part of the reason for this is that its capabilities and limitations haven't been fully explored.
As a result, while many of the problems may be similar, recognizing them and knowing what to do when you find them remains challenging, even to experienced Java developers. What follows are just a few of the development issues and what they mean for Java.
Performance is a concern of applications written in any language. Most programmers are familiar with common performance issues using a conventional language such as C with a stand-alone or even a client/server application. Such issues often involve improper allocation, deallocation of memory and poor use of system APIs.
These aren't even characteristics of the Java language. For example, relating Java code to how the JVM manages memory is difficult and error-prone. However, it's vital to improve Java performance because its execution model has additional overhead that tends to degrade performance more than native applications do.
In addition to typical performance requirements and issues surrounding traditional stand-alone applications, distributed Java applications must contend with problems surrounding the interactions between components running on different systems. Performance problems may manifest themselves in unexpected ways or appear to be caused by different parts of the code other than the actual problem area. Identifying and locating performance bottlenecks rapidly is a significant challenge in distributed application development.
Enterprise Java applications, especially distributed ones, are often mission-critical in nature: all aspects of the application must work perfectly at all times. Developers and development managers must be able to gauge the reliability of their applications accurately. While the characteristics of the language tend to make Java applications less error-prone, there are still plenty of ways to introduce runtime errors.
With distributed Java applications the reliability equation is even more difficult. It means assessing not only the individual applications but also the components as they interact. Java also makes it possible to write highly threaded applications that make sense in a distributed environment. But using threads means that problems with resource contention and deadlock are much greater.
Java applications face the same testing problems as traditional ones. They must be tested as thoroughly as possible before fielded, and developers should have a good idea of the extent of the test coverage before certifying an application. Distributed software systems written in Java, however, are extraordinarily difficult to test and debug. Because components reside on different computers and must work together perfectly for the application to work properly, all components must be tested simultaneously.
Since JVMs perform all the memory management tasks for applications, understanding the memory usage of the underlying platform and influencing memory allocation and use to affect performance is difficult to do and not intuitive. That's especially true because developers have no easy way of determining the relationship between code and the underlying memory use. Analyzing underlying memory use is a key component of building efficient applications (see Figure 1).
Software Tools Can Make Java Transparent
Several of the integrated development environments for Java are quite good, combining visual development, context-sensitive editing, JIT compilation and runtime debugging. What they lack, however, is the ability to determine the efficiency and reliability of the application, especially if it's distributed across several servers.
Most developers do without such tools, due in part to the relatively poor selection. There are many different Java-oriented development environments, but few tools to move code beyond the development stage. A few tools, such as the Compuware DevPartner for Java suite, combine components that evaluate performance issues, examine memory usage, analyze threads and track testing progress (see Figure 2).
Many Java developers fail to recognize that it's not enough for Java applications to be debugged within the development environment because of several myths regarding the use of Java as a development language and environment. One myth is that the VM eliminates programming errors and bugs. While direct memory errors aren't usually possible in Java development, it's still possible for Java applications to contain serious errors that affect the proper operation of the program.
Another myth is that the application developer has no control over the performance of the software since the VM manages the low-level details that determine how fast the code runs. How developers use specific language instructions can have a significant impact on application performance. Often a few simple changes can greatly improve performance if developers know their underlying effect in advance.
Java development tools assist and accelerate the development of reliable, high-performance applications, especially distributed ones. They go beyond the traditional development environments to include components that help make applications more reliable and efficient. When choosing a set of tools to supplement those found in development environments and improve the reliability and performance of Java applications, developers and development teams should take the following characteristics into account.
- Support for multiple, unmodified VMs: Different operating systems often use different JVMs with different performance and behavior characteristics. In addition, some Java tools require the use of specially modified VMs that may not represent the characteristics of production systems.
To obtain accurate information to improve the performance and reliability of distributed Java and mixed-language applications, developers should select tools that run in the actual deployment environment. This way they can be certain that the behavior observed and analyzed during development and testing will be the same once the application is deployed.
- Minimal impact on Java runtime resources: During application testing and analysis, the Java development tools that are used can often be intrusive - their presence may influence the test results. If developers attempt to deploy the tools for further testing and tuning in the production environment, they may also encounter unacceptable performance degradation due to high resource utilization.
If developers need accurate information on resource utilization, performance and system load, they should consider software tools that don't use large amounts of Java runtime resources. Tools with minimal Java and computing resource impact are more likely to provide the accurate information needed to improve the performance and reliability of distributed applications.
Application developers need tools that work with whatever software technologies they're using. Multilanguage tools, especially those that support industry-standard Web languages, provide the range of capabilities needed to find performance and reliability problems anywhere in the application (see Figure 3).
- Ability to track Java memory utilization to the function and line level: Knowing you have a performance or reliability problem isn't any good unless you're able to pinpoint the source rapidly and easily. The more precise the diagnosis and analysis, the faster application developers can address the problem.
Java application development tools should focus as specifically as possible on the exact location of a performance bottleneck or software error, especially for large, distributed applications. Developers working on deadlines need the most exacting information possible from their tools.
- Ability to find thrashing and starvation conditions graphically: Some of the toughest software problems to find are those that involve resource starvation and code thrashing. These error conditions don't prevent the application from running; however, they cause severe performance bottlenecks and may even cause the application to hang while running.
The problems are especially prevalent in Java applications in which multiple running threads contend for limited virtual machine resources. To ensure the reliability of deployed applications, Java developers require software tools that enable them to identify and locate the complex combination of conditions that can cause resource starvation and thrashing.
- Ability to measure code base stability: Rapidly changing code during debugging and testing usually means that the application may be unreliable or needs additional testing before deployment. Conversely, an application with few changes to the code base during debugging and testing will more likely be fully tested and stable.
It's important, therefore, for application developers to understand how much and how rapidly their application code is changing during the latter stages of the development process. An application development tool should measure the stability of the code base to enable both the development team and the managers to determine when the application can be used reliably in production.
- Support for multiple operating systems: Mixed operating environments are the rule rather than the exception. Even if developers code on one platform, it's increasingly likely that the application will be deployed across several different types of systems. This is especially true of Java-based applications, which are designed to run unmodified on multiple operating systems.
Application development tools have to support multiple development and runtime environments to eliminate the expense of purchasing different tool sets for different platforms, and to reduce the need for developer training on multiple tool sets and platforms.
Delivering Higher Quality Java Applications
Today's Web-enabled, distributed applications combine many different technologies and are prone to performance and reliability problems. Software developers using Java technology can spend a substantial amount of time trying to resolve these problems, leading to schedule delays and applications with ongoing problems.
Most of the Java development environments available are excellent for writing small to medium-sized stand-alone applications. It's growing increasingly difficult to write applications that are large or distributed, and to work with legacy components or databases. Performance and reliability issues overshadow the advantages of rapid, object-oriented development.
Visualizing these types of problems is an important aspect of debugging, tuning and testing applications because developers can quickly identify and localize the code responsible. If the developer can see where the deadlocked thread is, it's easier to pinpoint the resource that's deadlocked and its cause.
The kinds of problems that don't arise or are trivial in smaller applications take on critical importance in distributed processing. By using performance analyzers, memory profilers and thread inspectors such as those found in DevPartner Java Edition, distributed application developers can be sure their applications will run as expected.
Large-scale Java and distributed applications are easy to write but difficult to write well. To bridge the gap between mediocre or poorly performing applications and highly efficient ones, developers need to leverage software tools that help build reliable high-performance applications and components with Java technology. Such tools would allow Java developers to quickly and easily identify problems in key areas like runtime performance, memory utilization and multithreading.
Peter Varhol is a technical evangelist for the
Compuware NuMega product line. He holds graduate degrees in
computer science, applied mathematics and