NUMA or Non-Uniform Memory Access is a multiprocessing architecture that distributes memory access to different processors. Unlike SMP or Symmetric Multiprocessing, where several CPUs share a single memory, NUMA has a separate close and distant bank for memory. This architecture scales more effectively to more CPUs than SMP.
The term ‘non-uniform’ refers to the difference in speed between local memory and shared memory. With NUMA, memory on the same processor board as the CPU (local memory) can be accessed more quickly than memory on other processor boards (shared memory).
NUMA provides efficient and fast memory access to the processor, by enabling the processor to access the memory near to it quickly. This architecture is preferable for applications that have a large amount of data to access in parallel.
The phrase ‘Cache coherent NUMA’ indicates that the local system supports caching, which provides a speedy memory access to the processors.
What is NUMA architecture used for?
NUMA architecture is used to distribute memory access to different processors in a server system. It is ideal for large applications with parallel processing needs and provides efficient and fast memory access to the processors.
What is the difference between SMP and NUMA?
While both SMP and NUMA are multiprocessing architectures, the difference lies in how they access memory. SMP has several CPUs that share a single memory, while NUMA has separate close and distant banks for memory access. This architecture scales more effectively to more CPUs than SMP.
Why is NUMA architecture scalable?
NUMA architecture provides efficient and fast memory access to the processors by enabling the processor to access the memory near to it. This scalability makes it easier to add processing units to the system, making it ideal for large applications with parallel processing needs.
NUMA or Non-Uniform Memory Access is a multiprocessing architecture preferred for large applications with parallel processing needs. Unlike SMP, NUMA provides a separate bank for memory access and enables faster memory access near to the processor. This makes it a scalable architecture ideal for server systems that need to add processing units with ease.