Flynn's Taxonomy

What is it?

Michael J. Flynn, a Stanford’s computer scientist, proposed in 1966, a classification of all computer architectures, based on the instructions stream and data flow, which is known as Flynn’s Taxonomy. It defines four categories: SISD, MISD, SIMD, and MIMD.

• SISD - Single Instruction, Single Data

  Consists in a single processor which runs instructions one at a time, storing data in a single memory. Instructions are retrieved from memory, processed, and then returned to memory. e.g. Von Neumann Architecture

• SIMD - Single Instruction, Multiple Data

  Consists in multiple processing units which run the same instruction simultaneously, with each processing unit with its own data set, obtaining different results for the same instruction. e.g. GPUs fall into this category A heavy workload is divided into multiple smaller pieces, with each processing unit taking a small piece. The results are later re-assembled. MapReduce works similarly.

• MISD - Multiple Instruction, Single Data

  MISD systems are the least used, once MIMD and SIMD are more flexible and scalable. However, when running a pipeline of instructions, MISD systems are extremely resistant to partition and failures, where nodes validate each other results. e.g. NASA used it for flight-control algorithms

• MIMD - Multiple Instruction, Multiple Data

  Consists in multiple processors running different instructions in different data, at the same time, asynchronously and independently. These can be divided into two groups: shared memory and distributed memory.

  • Distributed Memory

    Each processor has its own memory and runs independently. The communication between processors are made in a external bus. i.e. Clusters and horizontal scaling

  • Shared Memory

    Multiple processors access the same memory, normally using SMP (Symmetric Multi Processing) to manage memory. This is the architecture used in most devices today. with uniform access to memory for all processors. However, a recent architecture with non-uniform access to memory has been created. NUMA (Non-Uniform Memory Access) is normally used in servers, which is basically a cluster of multiple SMP nodes.