Skip to content

Oversubscription

Category: Foundations · Status: stub

One line: More busy threads than there are cores, so the machine spends its time switching between them instead of running them.

How it connects

flowchart LR
  n_contention["Contention"]
  n_oversubscription["Oversubscription"]
  n_oversubscription -->|can cause| n_contention
  classDef center stroke-width:3px
  class n_oversubscription center
  classDef outside stroke-dasharray: 4 3
  class n_contention outside

In each language
Rust available_parallelism ↗ estimates how many threads a program should use
Go runtime.GOMAXPROCS ↗ caps how many CPUs execute Go code simultaneously, however many goroutines there are
C++ std::thread::hardware_concurrency ↗ is only a hint, and may be 0
Python os.process_cpu_count ↗ (3.13) counts the logical CPUs the calling thread may use, and ThreadPoolExecutor ↗ sizes its default pool from it
JavaScript navigator.hardwareConcurrency ↗ is the number of logical processors available to run threads
Kotlin Dispatchers.Default ↗ uses at most as many threads as there are CPU cores, and at least two
The operating system sched_getaffinity ↗ gives the CPUs a thread may run on, which can be fewer than the machine has

Where to read more

  • In the books: Pro TBB, Michael Voss, Rafael Asenjo, James Reinders — ch. 11, 'Controlling the Number of Threads Used for Execution'
  • In the books: C++ Concurrency in Action, Anthony Williams — ch. 2, 'Managing threads' → 'Choosing the number of threads at runtime'
  • In the books: Functional and Concurrent Programming, Michel Charpentier — ch. 24, 'Case Study: Parallel Execution' → 'Bounded Number of Threads'
  • Notes: oversubscription ↗