Chapter 2-- Parallel Architectures

CPU(Central Processing Unit) -- 中央处理器

GPU( Graphic Processing Unit ) -- 图形处理器

VLSI(A circuit containing one hundred thousand to one million electronic units on a chip) -- 超大规模集成电路

Commodity cluster -- 常用集群  ASCI -- 先进计算计划  network of workstations -- 工作站网络

processor arrays -- 处理器阵列  multiprocessors -- 多处理器 multicomputers -- 多计算机

systolic arrays -- 脉动式阵列

Interconnection Networks -- 互连网络: shared memory or send message to each other

Shared versus Switched Media -- 共享介质和开关介质

A shared medium allows only one message at a time to be sent. Processors broadcast their message  over the medium. Each processor "listens" to every message and receives the one for which it is the destination.(共享介质允许在某一时刻仅仅一个消息被发送，通过共享介质，处理器得以广播它们的消息，每个处理器监听每一条消息并接收发送至其上的消息)

Switched interconnection media support point-to-point message among pairs of processors. Each processor has its own communication path to the switch. Switches have two important advantages over shared media. They support the concurrent transmission of multiple message among different pairs of processors, and they support the scaling of the interconnection network to accommodate greater numbers of procrssors.(开关互连介质支持处理器对之间点对点的消息互传，每个处理器都有属于它自己的通向开关通信的一条路径。与共享介质相比，开关介质有两个重要的优势，它们支持在不同处理器对之间的多个消息并发传输，而且支持互连网络的扩展，从而能够容纳更多数目的处理器)

Switch Network Topologies 开关网络拓扑结构

vertices = processors or switches

edges = communication paths

type ：

Direct

    Ratio of swtich nodes to processor node is 1：1

    Every switch node is connected to

            * 1 processor node

            * at least 1 other switch node

Indirect Topology

  Ratio of switch to processor nodes is greater than 1：1

  Some switches simply comnect other switches.

Evaluating Switch Topology -- 评价开关网络拓扑

* Diameter -- the maximum distance

* Bisection Width -- 对分带宽

* Number of edge node -- 每个开关节点的变数

* Constant edge length -- 固定边长