dragonwell-project/dragonwell8
Dragonwell8
Dragonwell8 is an OpenJDK distribution and Java Virtual Machine designed for high-throughput big data processing and large-scale cloud deployments. It functions as a big data runtime and JIT compilation optimizer, featuring a coroutine-based threading model and dynamic heap memory reclamation to reduce system overhead.
The project distinguishes itself through native acceleration libraries and RDMA-based network providers optimized for Spark workloads and large-scale data processing. It further reduces application startup times and eliminates initial performance dips using profile-guided JIT warmup and a shared class data architecture.
The runtime includes a comprehensive diagnostic toolset for real-time event streaming, selective heap snapshotting, and object allocation profiling. Additional capabilities cover resource isolation management to control CPU and memory usage across different tenants and performance tuning for high-concurrency asynchronous tasks.
Features
- JDK 8 Runtime Compatibilities - Runs unmodified JDK 8 applications with AI-extension capabilities, requiring no code adaptation or migration.
- Native Acceleration Libraries - Loads scenario-specific acceleration libraries to speed up Apache Spark jobs on x86-64 platforms without code changes.
- Native Library Accelerations - Loads scenario-specific acceleration libraries to speed up Apache Spark 3.5.x jobs on x86-64 platforms without code changes.
- Native Spark Accelerators - Loads native acceleration libraries and RDMA-based network providers to speed up Apache Spark jobs without code changes.
- Class Data Caches - Records class loading data during a first run and replays it on subsequent runs to reduce startup time.
- OpenJDK Distributions - An OpenJDK distribution and Java Virtual Machine optimized for high-throughput big data processing and large-scale cloud environments.
- Big Data Runtimes - A Java runtime with coroutine-based threading and dynamic heap reclamation designed to reduce overhead in big data workloads.
- Profile-Guided JIT Compilers - Uses profile-guided warmup and shared class data to accelerate application startup and eliminate performance dips.
- Profile-Guided Warmup - Records and replays class-loading and hot-method profiles at startup to accelerate application warmup and reduce cold-start latency.
- JVM Coroutine Implementations - Converts Java threads into lightweight coroutines to improve concurrency and reduce context-switching overhead.
- Thread-to-Coroutine Converters - Converts selected Java threads into lightweight coroutines to improve concurrency and reduce context-switching overhead.
- Shared Class Data Archives - Dumps loaded classes into a shared file so later JVM instances can skip re-loading them from disk.
- Big Data Runtime Optimizations - Optimizes BigDecimal arithmetic and Spark TPC-DS performance for AI and big data scenarios, boosting throughput by over 5%.
- Dynamic Heap Reclamation - Dynamically returns unused heap memory from G1 to the operating system, lowering the Java process's memory consumption.
- Startup Cache Mode Selections - Selects among trace, profile, dump, replay, or destroy modes to manage how startup acceleration data is collected and applied.
- I/O-Aware Garbage Collections - Adjusts GC behavior under heavy I/O loads to reduce pause times and improve application responsiveness.
- Multi-Tenant Resource Isolation - Controls CPU and memory usage per tenant, enforcing resource quotas and data isolation within a single JVM.
- JVM Multi-Tenant Resource Controllers - Controls CPU and memory usage per tenant within a single JVM for secure multi-tenant deployments.
- JVM Multi-Tenant Resource Controllers - Controls CPU and memory usage per tenant, enforcing resource quotas and data isolation within a single JVM.
- JVM Diagnostic - Ships a comprehensive toolset for real-time event streaming, selective heap snapshotting, and object allocation profiling in the JVM.
- JVM RDMA Providers - Uses Remote Direct Memory Access for high-throughput data transfer in large-scale data processing workloads.
- JFR Event Streams - Streams Java Flight Recorder events continuously for live monitoring without dumping a recording file.
- Object Allocation Profilers - Samples object and array allocations in the C2 compiler and emits dedicated JFR events to trace hot allocation paths.
Star history
How this analysis was created: This summary and feature list are AI-generated from collected project material and can contain mistakes. Stars, license and language are imported from GitHub. Inclusion does not mean that we have tested or audited this project. Check the source documentation for any feature you depend on. Learn more on our About page.
