How to Calculate the Optimal Thread Pool Size for Java Applications

Problem Statement

How to calculate the number of concurrent threads in a Java thread pool is debated in two well‑known books. The article presents the two formulas and investigates which one is correct.

Formula from "Java Concurrency in Practice"

The book gives: Nthreads = Ncpu * Ucpu * (1 + w/c) Ncpu : number of CPU cores

Ucpu : CPU utilization (0‑1)

w/c : ratio of wait time to compute time

Formula from "Programming Concurrency on the JVM"

It proposes: Nthreads = Ncpu / (1 - blocking_factor) where blocking_factor = w / (w + c) (wait time / (wait + compute)).

Analysis

Assuming 100 % CPU usage (Ucpu = 1), the first formula simplifies to Ncpu * (1 + w/c). Equating both formulas yields blocking_factor = w/(w + c), confirming the two expressions are mathematically equivalent.

Practical Application

For I/O‑bound tasks (w/c > 1) a conservative choice is Nthreads = 2 * Ncpu. For CPU‑bound tasks (w = 0) the optimal size is Nthreads = Ncpu. In Java you can obtain Ncpu via Runtime.getRuntime().availableProcessors().

IO‑bound = 2 × Ncpu (generally safe); CPU‑bound = Ncpu (or Ncpu + 1 for a slight safety margin).

Conclusion

Choosing a thread‑pool size depends on task type, memory constraints, and overall resource usage; the two formulas ultimately describe the same relationship.