Multiprocessors

Multiprocessors are computer systems that contain two or more central processing units (CPUs) capable of executing instructions simultaneously. In such systems, the processors share memory and I/O resources, enabling higher throughput and parallel computation compared to single-processor machines. Multiprocessors are commonly categorized by how memory is shared: in symmetric multiprocessing (SMP) systems, all processors access a single shared address space through a common interconnect; in distributed-memory multiprocessors, each processor has its own local memory and processors communicate by passing messages.

In SMP, cache coherence is essential because each processor maintains local caches; coherence protocols such as MESI ensure that reads and writes observe a consistent view of memory. Performance depends on memory bandwidth and interconnect latency. NUMA (non-uniform memory access) architectures allow processors to have faster access to nearby memory banks while presenting a single system image; NUMA-aware operating systems and schedulers help place threads to maximize locality.

Common interconnects include buses, crossbar switches, and multi-dimensional mesh networks. The choice of interconnect affects scalability, contention, and fault tolerance; large SMP systems use sophisticated networks to reduce bottlenecks.

Software and programming: parallel programming models, multithreading, and parallel libraries; Amdahl's law highlights limits to speedups due to non-parallelizable portions. Operating systems provide multiprocessing support, thread scheduling, and synchronization primitives; developers must manage data sharing and locking to avoid race conditions and deadlocks.

Usage and trends: Multiprocessors are found in servers, workstations, and high-performance computing systems. Modern general-purpose processors increasingly integrate multiple cores on one chip, blurring the line between multiprocessors and multi-core architectures, but most large systems still rely on multiple physical CPUs to deliver high throughput.