Symmetric Multiprocessing
Subject: Computer and Technology noun /sɪˈmɛt.rɪk ˌmʌl.tiˈprɑː.sɛs.ɪŋ/ 1 definition
- computer-architecture
- operating-systems
- parallel-computing
-
Definition (noun)
Entry #200 · UpdatedSymmetric Multiprocessing (SMP) is a multiprocessor architecture in which two or more identical processors connect to a single, shared main memory and operate as equals under one operating system instance. Unlike asymmetric multiprocessing—where a primary processor delegates specific tasks to subordinate processors—SMP lets any CPU execute kernel routines, service interrupts, and schedule threads from a common work queue. By presenting a unified, cache-coherent memory space across all processors, SMP simplifies parallel programming and enables dynamic load balancing, though throughput can eventually be bounded by memory bus contention and cache-coherency overhead as core counts grow.
Common Forms
Uniform Memory Access (UMA) SMP: Connects all processors to a centralized memory pool over a shared bus or crossbar switch, giving every CPU equal latency and bandwidth to any memory address.
Cache-Coherent NUMA (ccNUMA) SMP: Distributes physical memory across processor sockets while hardware coherence protocols preserve one unified address space, scaling to higher core counts despite faster access to local memory than remote memory.
Multi-Core (Chip-Level) SMP: Integrates multiple identical execution cores onto a single die or package with shared last-level caches, lowering inter-processor communication latency compared to multi-socket boards.
Virtual SMP (vSMP): Allocates multiple virtual CPUs from a hypervisor to a single guest virtual machine, enabling an unmodified operating system to schedule parallel workloads across physical host cores.
Example
A database server equipped with 16 identical CPU cores and 128 GB of shared RAM runs a single Linux kernel that manages all cores as equal peers in an SMP configuration. During a traffic spike, the OS scheduler dynamically assigns incoming query worker threads to idle cores while every thread reads and updates the same in-memory buffer pool.