Day 1: computer architecture classification, part of memory

Mainly include Flynn classification and Feng's classification.

The Flynn classification is mainly based on different combinations of command streams and data streams.

Concepts:
Command Stream: Instruction Stream, the sequence of commands executed by machines
Data Stream: Data Stream, a sequence of Data for instructions, including input Data and intermediate result Data
Multiple: Multiplicity: the maximum number of commands or data that are in the same stage of execution on components that are equivalent to system bottlenecks (this data is not used in the classification and I don't know what to do)

Combination is simple, 2*2 = 4
Single command stream single data stream (SISD)
Single command stream multi-data stream (SIMD)
Multi-instruction flow, single data stream (MISD)
Multi-command stream multi-data stream (MIMD)

SISD is easy to understand. Isn't SIMD equivalent to a bunch of prisoners queuing up, and then the execution of one by one shooting in order means almost the same. Some people say that because data is controlled by commands, MISD does not actually exist. Some people also say that "the fourth type is the so-called Multiple Instruction Single Data computer. In this computer, each processing unit forms a linear array and executes different command streams separately, while the same data stream passes through each processing unit in this array sequentially. This system structure applies only to some specific algorithms .", I cannot understand. Is MIMD equivalent to putting a bunch of sisds together?

The most difficult thing is the Feng's classification. The formula for the average degree of parallelism and the average utilization is better understood, mainly as follows.

First, what is a Word )? What is Bit )? What is corresponding to the real system? If you understand the two, the rest will be simple. The horizontal axis is N, indicating the number of digits simultaneously processed in a word, and the vertical axis is M, indicating the number of words simultaneously processed in a bit. Depending on the two situations, there can be four types:
String WSBS, n = 1, m = 1
String with WSBP, n> 1, m = 1
String WPBS, n = 1, m> 1
Word merging and WPBP, n> 1, m> 1

The difference between computer system structure and computer composition: the former determines what functions are needed, and the latter determines how to implement this function.

Parallelism includes two concepts: synchronicity and concurrency. Synchronization refers to the occurrence of two or more events at the same time. Concurrency refers to the continuous occurrence of two or more events at the same time interval.

Memory Hierarchy: high-speed cache, primary storage, and secondary storage. Some machines do not have high-speed cache, so they become two layers.
Memory classification is messy. There are many different types of storage according to different conditions.
Memory and external memory by location,
By material, there are many magnetic, semiconductor, and optical
According to the working method, Read Only Memory is available, that is, ROM; Programmable Read Only Memory. The manufacturer can write it once, and then it cannot be written again. Erasable Programmable Read Only Memory, which can be erased by users. However, it takes several minutes to clean the file. Flash Memory is not a USB Flash drive. Although Flash Memory is faster, it is not a USB Flash drive.
Access by address or by content
According to the addressing method, there is random memory (as if the memory we use belongs to this), sequential memory (the previous tape or something), direct memory (in fact, it is in the first two types)

The following describes an "connected storage". I don't know what it means to put it here. It indicates that it is a type of memory accessed by content. it searches for a piece of content in each storage unit and finds all the data words that are the same as the content. In computer systems, the connected memory is mainly used to store segment tables, page tables, and express tables in virtual memory. In the cache, the connected memory is used as the row address for storing cache. This is because you need to quickly search for these two applications.

I know the size of the comparator, what bit comparator, word comparator, 2 m * N, 2 m, and N are bit widths, but why is it 2 m? It's big. Continue tomorrow, cache ....