Why LongAdder Beats AtomicLong for High‑Concurrency Counters in Java

1. Introduction

In distributed systems, a counter is a common requirement. To achieve high concurrency and high availability, we need to choose an appropriate implementation.

In Java, the two typical counter implementations are AtomicLong and LongAdder .

Alibaba’s technical report recommends using LongAdder instead of AtomicLong .

This article introduces the principles, advantages, and disadvantages of both counters and analyzes why Alibaba prefers LongAdder.

2. CAS

2.1 Full Name

CAS stands for compare and swap , an atomic operation corresponding to the CPU instruction cmpxchg.

2.2 Simple Understanding

CAS has three operands: current value A, memory value V, and new value B.

If A equals V, the memory value is changed to B.

If A does not equal V, the operation retries or aborts.

The core of CAS is comparing the current value with the memory value to detect modifications.

2.3 Issues of CAS

CAS suffers from the ABA problem.

Example:

Thread A reads value 10.

Thread B changes it to 100.

Thread C changes it back to 10.

When thread A resumes, it sees the value unchanged and proceeds, although the value was modified twice from a global perspective.

2.4 Solving the ABA Problem

Java provides AtomicStampedReference, which adds a version stamp to the value, so the comparison includes both the value and its version.

3. LongAdder

3.1 What Is LongAdder

LongAdder, introduced in JDK 1.8 by Doug Lea, resides in java.util.concurrent.atomic. It offers better performance than AtomicLong in high‑concurrency scenarios at the cost of higher memory usage.

LongAdder is a high‑performance counter from Google’s open‑source projects. It splits a long into multiple 16‑byte cells, each updated by an independent AtomicLong, allowing threads to update different cells concurrently.

Advantages: high concurrency performance, scalability, and reduced contention. Disadvantage: more complex implementation and higher memory consumption.

3.2 Why LongAdder Is Recommended

LongAdder reduces contention by using segmented locks. It maintains a base value and an array of cells, as illustrated below: