OpenJDK’s Project Leyden will implement condensers to improve the startup time, time to peak performance, and resource footprint of Java programs.
Project Leyden, an ambitious effort to improve startup time, performance, and footprint of Java programs, is set to offer condensers. A condenser is code that runs between compile time and run time and transforms the original program into a new, faster, and potentially smaller program.
In an online paper published October 13, Mark Reinhold, chief architect of the Java platform group at Oracle, said a program’s startup and warmup times and footprint could be improved by temporarily shifting some of its computation to a point either later in run time or backward to a point earlier than run time. Performance could further be boosted by constraining some computation related to Java’s dynamic features, such as class loading, class redefinition, and reflection, thus enabling better code analysis and even more optimization.
Project Leyden will implement these shifting, constraining, and optimizating transformations as condensers, Reinhold said. Also, new language features will be investigated to allow developers to shift computation themselves, enabling further condensation. However, the Java Platform Specification will need to evolve to support these transformations. The JDK’s tools and formats for code artifacts such as JAR files will also need to be extended to support condensers.
The condensation model offers developers greater flexibility, Reinhold said. Developers can choose which condensers to apply and in so doing choose whether and how to accept constraints that limit Java’s natural dynamism. The condensation model also gives Java implementations considerable freedom. As long as a condenser preserves program meaning and does not impose constraints except those accepted by the developer, an implementation will have wide latitude for optimizing the result.
To improve startup and warmup time and footprint can be best done by identifying computation that can simply be eliminated, Reinhold said. Failing that, computation can be shifted backward or forward in time. This concept of shifting computation in time is not new. Java implementations already have many features to shift computation. For example, compile-time constant folding shifts computation backward in time from run time to compile time, and garbage collection shifts the reclamation of memory forward in time. Other computation-shifting mechanisms are optional including ahead-of-time computation and class-data sharing.
Project Leyden was under discussion for more than two years before beginning to move forward earlier this year. The project is sponsored by the HotSpot virtual machine and core libraries groups within the Java development domain.
