Shulou(Shulou.com)06/01 Report--

In 1966, MichealFlynn classified the computer architecture according to the concepts of instruction and data flow, which is called Flynn classification. Flynn divides computers into four basic types, namely SISD, MIMD, SIMD, and MISD.

The traditional sequential computer can only execute one instruction (that is, only one control flow) and process one data stream (that is, only one data stream) at the same time, so it is called single instruction single data stream computer Single Instruction Single Data or SISD computer). For most parallel computers, multiple processing units perform different operations and deal with different data according to different control flows. therefore, they are called multi-instruction stream and multi-data stream computers, namely MIMD (Multiple Instruction Multiple Data) computers.

Vector computers, which have become the mainstream of super parallel computers for a long time, in addition to scalar processing units, the most important thing is to have hardware units that can perform vector computing. When performing vector operations, an instruction can operate on multiple data (forming a vector) at the same time, which is the concept of single instruction stream multiple data stream (Single Instruction Multiple Data,SIMD). Therefore, we call vector computers SIMD computers.

The fourth type is the so-called multi-instruction stream single data (MultipleInstructionSingleData) computer. In this kind of computer, each processing unit forms a linear array and executes different instruction streams, while the same data stream passes through each processing unit in the array in sequence. This system structure is only applicable to some specific algorithms.

Relatively speaking, SIMD and MISD models are more suitable for special computing. In commercial parallel computers, the MIMD model is the most general, SIMD is the second, and MISD is the least used.

Multiprocessor. A parallel processor. Pipeline processor

1. With regard to "parallel processing", we should understand: a. Objective: to save time in solving large and complex problems. b. Features: processors can work on different aspects of the same program simultaneously. In order to use parallel processing, you first need to: (see c) c. Prerequisite: parallelization of the program, that is, each part of the work is assigned to different processors. d. Note: parallelism is a problem of interdependence and cannot be implemented automatically. Therefore, parallelism cannot guarantee acceleration. But: (see e) e. Advantage: a program executed on n processors may be n times faster than that executed on a single processor. two。 Array processor versus multiprocessor: array processor (parallel processor): single control unit (CU), multiple processing units (PU); shared main memory (which can be divided into multiple) and all external devices; single instruction stream multiple data stream computer (SIMD). Parallelism at the operation level is realized, and parallelism exists within the instruction. Multiprocessor: multiple control units, multiple processors (each processor has its own control unit); sharing one main memory and all external devices; multiple instruction stream multiple data stream computer (MIMD). Achieve job, task-level parallelism. Parallelism also exists outside the instruction. 3. Comparison between array processor and single-machine system with pipelined structure: array processor: single instruction stream multiple data stream; using resource repetition technology. The stand-alone system with pipelined structure: single instruction stream and multiple data streams; using time repetition technology. 4. Multiprocessor: a multiprocessor system consists of multiple independent processors, each of which can process its own program independently. Classified according to its composition: heterogeneous (asymmetric): composed of multiple processors of different types or capable of performing different functions; isomorphic (symmetrical): composed of multiple processors of the same type or capable of performing the same function; distributed processor system: multiple processors with independent functions are connected to each other. According to the interconnection structure between computers, it is divided into four kinds of multiprocessor structure: bus structure: single bus, multi-bus, hierarchical bus, ring bus and so on. Crossover switch structure: a set of vertical and horizontal switch arrays are arranged to connect the horizontal processor P and Istroke O channels with the vertical memory M. Multi-port memory structure: a plurality of corresponding ports of multi-port memory are connected together, and each port is connected to a processor P or an Imax O channel controller I/OC. Switch pivot structure: there are multiple inputs and outputs, switching between them, so that the input is selectively connected to the output. An arbitration unit (a component that can decompose conflicts) and a switch unit (which can be switched between one input and multiple outputs) form a basic switching hub. 5. Array (parallel) processor: two typical structures: parallel processor architecture with distributed memory: there are two types of memory: one is attached to the main processor and its resident operating system, and the other is local memory, distributed in various processing units. used to store programs and data. Parallel processor architecture with shared memory: several shared memories are shared among all the processing units of the whole parallel system through the internetwork ICN.

2009-05-11

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.