Original articles dissecting source code and sharing personal tech insights. A modest space for serious discussion, free from noise and bureaucracy.
This article explores the design and implementation of time wheel algorithms in Netty and Kafka, comparing their single‑layer and multi‑layer approaches, analyzing performance trade‑offs, and detailing how these systems achieve O(1) scheduling for massive delayed tasks while avoiding empty ticks.
This article provides a comprehensive analysis of Netty's memory‑leak detection mechanism, covering the design principles, leak‑aware buffers, reference handling, sampling strategies, detection levels, and the internal models that enable accurate identification of unreleased native memory in both pooled and unpooled ByteBuf allocations.
This article adds a missing section on Netty 4.1.112.Final memory pool metrics, detailing the various Metrics classes for PooledByteBufAllocator, PoolArena, PoolSubpage, PoolChunkList, and PoolChunk, and includes code examples for each.
This article examines Netty 4.1.x’s ByteBuf reference‑counting mechanism, explains why reference counting was introduced, traces its original design, shows instruction‑level optimizations, reveals concurrency bugs in version 4.1.17, and details the clever even‑odd redesign that guarantees thread‑safe memory release while preserving high performance.
Exploring the differences between PhantomReference and WeakReference in Java’s ZGC, this article explains how ZGC’s concurrent phases and the should_drop method use distinct bitmap bits to determine object liveness, revealing why only PhantomReference reliably tracks object reclamation in certain finalization scenarios.
This article explains how the JVM creates and manages FinalReference objects, how they interact with the finalize() method, and how the ZGC and other GC phases discover, revive, and eventually reclaim objects, detailing the roles of ReferenceHandler, FinalizerThread, and related internal data structures.
This article explains how ZGC in OpenJDK 17 decides whether to reclaim SoftReference objects, detailing the unconditional clearing scenarios, the policies governing SoftReference lifetimes, and how the JVM quantifies memory pressure using heap size and the -XX:SoftRefLRUPolicyMSPerMB flag.
This article explains how Java's System.gc triggers full or concurrent garbage collection in NIO scenarios, detailing the JVM's handling of out‑of‑memory errors during memory‑mapped file operations and direct buffer allocations, and examines the behavior across various garbage collectors such as Serial, Parallel, G1, ZGC, and Shenandoah.
This article examines the internal read/write mechanisms of Java's FileChannel and MappedByteBuffer on Linux kernel 5.4, compares their performance through detailed source analysis and benchmarks, and explains why MappedByteBuffer often outperforms FileChannel for small I/O but can be overtaken for larger transfers due to page‑fault and dirty‑page handling.
This article explains the Linux kernel's mmap implementation, covering how virtual memory is allocated, the role of get_unmapped_area, mmap_region, overcommit policies, VMA merging, and the underlying data structures such as mm_struct and vm_area_struct.
