This article explains the principles, advantages, and drawbacks of CAS, LongAdder, and AtomicLong in Java, illustrating why Alibaba recommends LongAdder for high‑concurrency counter scenarios, how the ABA problem arises, and how AtomicStampedReference can mitigate it.
This article explains the principles, advantages, and drawbacks of Java's AtomicLong and LongAdder counters, describes the CAS operation and its ABA problem, and analyzes why Alibaba recommends LongAdder for high‑concurrency, high‑availability scenarios in distributed systems.
In high‑concurrency Java applications, LongAdder—introduced in JDK 8 and using partitioned cells to reduce contention—generally outperforms the single‑value AtomicLong, which relies on CAS and can cause CPU waste under heavy load, so Alibaba advises LongAdder for scalable distributed counters, though memory usage and workload specifics must be considered.
This article explains the principles of CAS, compares Java's AtomicLong and LongAdder counters, highlights the ABA problem, and shows why Alibaba recommends LongAdder for high‑concurrency scenarios due to its segmented lock design, better performance, scalability, and ease of use.
This article explains the concept of spin locks, compares them with non‑spin locks, discusses their advantages and drawbacks, and provides Java code examples—including an AtomicLong implementation and a custom re‑entrant spin lock—to illustrate how spin locks work in concurrent programming.
This article explains how Java's LongAdder works, compares its API and performance to AtomicLong, provides usage examples and a benchmark showing its superior throughput under high thread contention, while noting its higher memory cost and appropriate scenarios for adoption.
This article explains why LongAdder was introduced to overcome AtomicLong's CAS contention bottleneck, explores the design of Striped64 and its cells, and walks through key source code snippets of LongAdder, DoubleAdder, and the longAccumulate method, highlighting concurrency mechanisms and performance considerations.
This article explains how java.util.concurrent.atomic.LongAdder provides a more efficient counting mechanism than AtomicLong under high contention, discusses its trade‑offs in space and precision, presents JMH benchmark results, and walks through the internal implementation details such as cells, CAS loops, and hash‑based distribution.
