Multicore vs Multiprocessor Scheduling
You'll learn: How scheduling strategies differ between chips with multiple cores versus systems with completely separate CPUs.
Understanding the Hardware Distinction
Both multicores and multiprocessors give you parallelism, but their physical layouts create different scheduling challenges.
A multiprocessor system has multiple separate CPU chips, each with its own package, often its own memory controller, and physically distant from others on the motherboard. Think of this like having multiple separate kitchens in different buildings—each fully independent.
A multicore processor packs multiple execution cores onto a single chip. These cores typically share lower-level caches (like L3) and memory controllers. This is like having multiple cooking stations in the same kitchen—they share pantry space and appliances.
Why Schedulers Care
For multicores:
- Cores share cache, so keeping threads that access similar data on the same chip improves performance (cache affinity is crucial within the chip)
- Migration between cores on the same chip is relatively cheap
- Heat and power constraints affect all cores together
For multiprocessors:
- Each CPU has separate caches, so cache affinity matters less between processors
- Migration between separate CPUs is more expensive (different memory paths, cache invalidation)
- NUMA effects are stronger—memory access costs vary significantly
- Each processor can often be managed more independently
Modern schedulers like Linux's CFS use hierarchical scheduling domains: they balance load aggressively within a multicore chip but migrate threads between separate processors more conservatively, respecting the higher cost of cross-processor moves.
Key Takeaway: Multicore scheduling prioritizes cache sharing and thermal management within a chip, while multiprocessor scheduling emphasizes NUMA-awareness and minimizes expensive cross-chip migrations—both extending processor affinity concepts to match physical hardware topology.