I'm curious about Project Leyden and its goals. Right now, it's focused on improving startup and warm-up times through partial Ahead-of-Time (AoT) compilation and using metadata from previous runs. Given the rise of microservices and modular architectures, do you think Leyden will eventually provide a complete AoT compilation option for the Java Development Kit (JDK)? In the modern context, dynamic loading seems less crucial since deploying multiple microservices in the cloud is often simpler. I also recognize that while GraalVM exists, using it standalone can be complex without frameworks like Quarkus to manage the initial setup. I'm interested in how this might level Java with languages like Go regarding performance and efficiency.
5 Answers
Leyden's goal is to reduce the computation done at runtime, shifting things to build time for better startup times. Ideally, this should include tasks like parsing configuration files alongside AoT, which would be a positive step forward.
Java's late entry into the static binary game due to AoT compilation could cost the language dearly. With the cloud now often favoring frameworks like Go that can compile to binaries, Java needs to catch up quickly. The community at large may view the JVM as resource-heavy, so unless Leyden offers a solid AoT solution, we might see a drop in Java adoption.
Using tools like Quarkus makes it easier to optimize build time and startup efficiency, but it's still a challenge for many developers.
Many users genuinely prioritize ease of use over pure performance. They want better startup times, but the real value lies in reducing the workload and improving user satisfaction in the long run.
Performance matters too, especially when the bills come due. That's a significant part of the equation.
Historically, AoT in Java has been attempted with tools like GCJ and ExcelsiorJet, which worked fine for desktop apps. But ultimately, not much interest developed during those times. I don't see the industry shifting back in that direction; the winds seem to be against it now.
I think Jet lost its appeal after open-source alternatives like OpenJ9 and GraalVM appeared. Jet might still be around for embedded systems but in general, the popularity has faded.
If serverless computing continues its rise, the demand for improved startup times might force a change. A lot of users are looking for ways to optimize that.
I don't think Leyden is aiming for complete AoT. There are other projects, like GraalVM, pursuing what you're talking about here. Leyden seems more focused on making the Java footprint smaller within a closed environment, which is a step up from the current Java Module System.
Are you suggesting that GraalVM is a better comparison here? It does seem more directed at offering AoT options.
You're right about the sentiment; even if performance isn't the biggest flaw, it is vital for adoption.