Summary
Leading voices from the Java community recently joined Azul for Prime Time, a virtual event dedicated to innovations and strategies that power high-performance Java estates.
In this post you will learn:
- Proven Strategies to Optimize Java Workloads and Reduce Waste
- Cloud-native compilation improves JVM fleet restarts for Java performance and scale
- Collaboration between engineering and finance teams is critical for reducing cloud costs
- Use of Java in AI workflows is increasing, particularly in data preparation and feature engineering
Java is the engine behind mission-critical applications and end-user experiences. From faster response times to consistent performance, Java can have a significant impact on business results and cloud costs. Leading voices from the Java community recently joined Azul for Prime Time, a dynamic virtual event dedicated to innovations, strategies, and real-world insights that power high-performance Java estates.
Here’s an overview of the presentations.
Write Once, Deploy Everywhere: From Containers to Serverless
James Ward, Principal Developer Advocate, AWS
James discussed the “write once, deploy everywhere” approach for Spring applications, focusing on seamless deployment across local development, containerization, and serverless environments. He emphasized optimizing local development feedback loops, using incremental compilers and unit tests, and ensuring service consistency with Postgres. For containerization, he highlighted the use of Spring Boot’s built-in containerization features and build packs. For serverless, he introduced the Serverless Java Container for running Spring applications on AWS Lambda, demonstrating a smooth transition from local development to production environments without significant code changes.
Making JVM Fleet Restarts Easier for Java Performance & Scale Using Cloud-Native Compilation
Simon Ritter, Deputy CTO, Azul
Simon discussed improving JVM fleet restarts for Java performance and scale using cloud-native compilation. He explained the inefficiencies of JIT compilation, particularly in microservices, where each instance undergoes a warm-up process. Ritter introduced the cloud-native compiler, which decouples JIT compilation from JVMs, improving resource utilization and reducing warm-up times. A real-world example showed a 7-minute warm-up time reduced to 10 seconds using the cloud-native compiler, saving 37.5% of CPU resources. This approach allows for better optimization, caching, and dynamic scaling, benefiting large-scale microservice deployments.
Cut Java Cloud Costs; Proven Strategies to Optimize Java Workloads and Reduce Waste
Rob Martin, FinOps Principal, FinOps Foundation
John Stuart, VP of Cloud Operations, Security & IT, Azul
The session focused on strategies to optimize Java workloads and reduce cloud costs. Rob highlighted the importance of collaboration between engineering and finance teams, emphasizing shared decision-making and early optimization. John discussed practical steps, such as selecting high-performance Java distributions, citing MasterCard’s 50% infrastructure reduction and Workday’s 90% pause time improvement. Both stressed the significance of the FinOps framework, training, and community engagement for cost-effective and efficient Java workload management
Migrating Java Workloads to the Cloud
Matt McClernan, Head of Specialist SA, Compute, ANZ, AWS
The discussion focused on migrating Java workloads to the cloud, emphasizing AWS services and Azul’s Prime JDK. Matt highlighted that Java remains prevalent, with 9 million developers and expected 50-60% growth in cloud-based Java workloads over three years. Common migration challenges include perceived difficulty, prioritizing feature releases, and justifying business impact. The AWS Migration Acceleration Program (MAP) streamlines migrations, offering investments and tools. Azul’s Prime JDK, compatible with OpenJDK, offers up to 35% performance improvements on Graviton processors, reducing latency and optimizing CPU utilization. The combination of Prime with AWS services like Graviton provides superior price-performance and cost optimization.
Building AI Agents with Spring & MCP
Josh Long, Spring Developer Advocate, Broadcom
James Ward, Principal Developer Advocate, AWS
Josh Long and James Ward demonstrated building an AI agent using Spring, MCP, and AWS Bedrock. They used the Nova Pro model for inference and the Cohere model for embeddings. The agent, named Assistant, helps adopt dogs from a fictional agency, Pooch Palace. They set up a Postgres database with vector store capabilities and integrated it with the agent. The agent uses RAG retrieval for similarity searches and MCP for tool integration. They also implemented chat memory persistence and observability using Spring Boot Actuator. The code for this project is available on GitHub.
Java & AI, Moving from Experiments to Business Implementations
Frank Delporte, Senior Technical Writer, Azul
Frank discussed the transition from Java AI experiments to business implementations. Frank, a Java developer at Azul, highlighted the growing use of Java in AI workflows, particularly in data preparation and feature engineering. He explained the difference between AI and machine learning, emphasizing the latter’s role in training computers to make decisions based on data patterns. Frank introduced various Java libraries like DeepNets, Deep Java Library, and LangChain 4J for model training and use, noting their ease of integration with existing Java systems. He also compared Java’s efficiency with Python, citing a study showing Java uses less energy and costs less, advocating for Java’s suitability for AI applications.
Java Perf & Scale; Mastering Techniques for Efficient Applications
Pratik Patel, VP of Developer Advocacy, Azul
Pratik discussed techniques for optimizing Java applications. He highlighted the importance of keeping applications simple and avoiding over-engineering. Patel demonstrated using the IntelliJ IDEA profiler on the Spring Pet Clinic application to identify performance bottlenecks, such as the “find owner” method call taking 10% of the time. He also introduced JHiccup for identifying JVM pauses and discussed coding practices like using primitives over wrapper classes and string builders for efficiency. Patel emphasized the significance of right-sizing JVM heaps and using appropriate garbage collectors, particularly the Azul Zing JVM for large heap sizes and low latency.
Scalable Agents with Koog & Ktor in Kotlin
Simon Vergauwen, Developer Advocate, JetBrains
Simon discussed scaling agents with Koog and Ktor in Kotlin. He explained configuring applications with environment variables and using data classes to load configuration files. Simon detailed starting a server with Koog using the Netty engine and handling concurrency with suspending lambdas. He highlighted Koog’s support for YAML, JSON, and properties files and the need to enable features like SSE. Simon also covered building AI agents, using tools like Google Maps, and handling errors and caching to improve efficiency. He emphasized the importance of detailed descriptions and serializable types for accurate LLM responses and concluded by noting Kotlin’s growing use in the AI world.
Conclusion
Watch every presentation from Prime Time on our website, absolutely free. You can also learn more about high-performance Java platforms.
