计算机网络复习笔记第四部分--广域网协议（英文版本）

Wide Area Network 广域网

different networks may be connected at any level of OSI layers
physical layer: repeater
data link layer: bridge
network layer and upper layers: router/gateway

Network routing 网络路由:

• one node may send packets (i.e.即, a message that is de-assembled to fixed size units) to another which is not locally connected, but situated in throughout a city, a country, or the continent.
• network protocols search for the best route(s) (usually the cheapest and the fastest) between two points —> routing algorithms
• network routing is not straightforward：
  – environment in real network is changing dynamically
  – a good route may attract lots of traffic —> leading to congestion, or even deadlock

Congestion and Deadlock

Congestion

an excessive过度 buildup堆积 of packets at one or more network nodes caused by

• a failure of one or more network nodes/links
• the number of packets exceeding the network’s ability
• etc…
  It may cause a chain reaction because the congestion at one node hampers妨碍 its ability to receive packets from other nodes

Deadlock

congestion becomes so severe that nothing moves

• store-and -forward deadlock
• reassembly deadlock
  

Congestion Control

• packet elimination 数据包消除
  – if an excessive buildup of packets occurs, eliminate some of them
  – unlucky senders’ destroyed packets do not reach their destination: senders must determine the fate of their packets; adverse不利的 effect is limited (for most cases).

• flow control
  – flow control protocols may regulate规定 the number of packets to be transmitted between two nodes.
  – congestion can still occur, if too many packets are flowing into a node

• buffer allocation 缓冲区分配
  – a source node requests nodes on the route for reserving保留 a sufficient buffer space
  – if the request is rejected, a source node looks for another route

• choke packet
  – each node puts any link(s) to a warning state if its utilization exceed a certain level
  – for any packet ongoing不间断的 the link under a warning state, its source node is notified通知 by a choke packet 被阻塞包通知
  – a node, that has received a choke packet, reduces the amount of transmission to that link 接收到阻塞包的话就减少对那个link的发送

Network Routing

我们想要解决两个问题：

1. 如何Router1知道目的地是哪个（比如Router3）
2. 如何让Router1知道到Router3的最短路径

Routing table

• we wish to find the least cost route
• given a network, a routing table shows the associated cost for communicating with each destination node
• it does not specify具体说明 the entire route for each destination, but it only indicates the next node
  
  
• the cheapest route from A to F is A—>B—>E—>F

how can we find such route?

Centralised routing 集中式路由选择

• a single node maintains all interconnection information, which is then broadcasted to the rest of nodes
• simple, but a failure of the central node, or any link connected to it, may affect entire network.

集中式路由选择就是说有一个中心点，它保存了整个网络的路由信息。会广播给其他的节点。但是如果这个点坏了，或者link坏了，可能会影响整个网络。

Distributed routing 分布式路由选择

• no central control, but each node determines and maintains the routing information independently
  – a node finds its neighbors and calculates the associated costs
  – each neighbor repeats the same procedure in turn
• complex, but a failure of a node or a link may cause little disruption扰乱

分布式路由时每个节点会存储相邻路由表信息。效率比较高。复杂。但是一个点出了问题对整个网络的干扰比较小

Static routing 静态路由

• once routing information is determined, it will not be modified调整，改进.
• simple, but insensitive to changing condition

Adaptive routing 自适应路由

• each node updates routing information according to dynamically changing network condition
• pitfall陷阱: some packet may shuttle back and forth among several nodes, thus never making any progress towards its eventual destination 一些包可能会在多个节点中来回穿梭，因此无法朝着最终的目的地前进
• can provide the recent information for link cost, but high overhead开销 for updating routing information frequently.

Hierarchical Routing 分层路由

当有太多节点在同一个网络里的时候你很难决定一个路径。所以我们可以structure the network hierarchically

Some features of hierarchical routing:

• all nodes are divided into groups, called domains, each of which is separate and independent network
• a route between two nodes in a common domain is determined using the domain’s protocol
• each domain has a designated指定 node, referred to as a router or a gateway, that searches for route(s) between domains
• it may consist of multiple subdomains

The Internet uses a hierarchical structure

RIP (routing information protocol)

• routers in the network let each other know the shortest route to a specified network
• typically, it uses a hop count (the number of intermediate中间的 routers) to measure the distance.
  
  Procedure：
  1） R1 informs N2它可以通过1 hop到达N1
  2）R2连接着N2所以它现在也知道自己可以两跳到N1，存储在routing table中
  3）R2向N3广播自己可以一跳到N2和N4，两跳到N1
  4）R3存储从N3获取的信息
  5）R3向N5广播自己可以一跳到N3，两跳到N2和N4，三跳到N1
  通过不停地重复这个过程，最终N6会知道自己可以通过N4三跳到N1，通过N5四跳到N1。

Dijkstra’s Algorithm 狄克斯特拉算法

centralised forward search algorithm 集中前向搜索算法

1. 定义S为一个节点的集合。最初，S里只包含一个A节点。
2. 对每一个在S外的节点node X，定义一个cost(X)作为仅使用S中的节点从A到X的最便宜路径的花销。
3. 定义一个prior(X)作为最便宜的路径中，X之前(preceding)的那个节点。如果根本不存在这样的link，那么cost(X)是一个任意大的值(arbitrarily large value)
4. 定义一个集合W，包含所有的不在S里，但是却和S里的节点有直接连接的节点。（其实就是邻居）
5. 从W里选择一个cost(X)最小的节点X。将这个节点X添加到S中
6. 只要还有节点不在S里，就重复从2开始的步骤
   

1）首先将A作为最初始的点，也是集合S的第一个元素。与A相邻的只有B和C，X选择B和C之中cost比较小的C。prior填写为B和C元素前面的那个元素。
2）然后C点作为上一轮的X，加入集合S。此时能与A，C有直接连接的B，D，E都进入集合W。cost计算应当是经过点C所计算出的cost。从A到点B的cost最低，加入X
3）接下来，X变成A，B，C。相邻的则有D，E和F。计算最短路径的时候可以计算经过B的和经过C的两种。这两种选择最低的作为cost。计算完后选择，如果有cost是一样的，随机选择一个。
4）按照这样的方式不停进行。直到所有的点都被收进S集合。例题中的最后一行就是最终结果。

Bellman-Ford Algorithm 贝尔曼-福特算法

nodeA的邻居们如何知道自己到Z的最便宜的路径呢? ----> Bellman-Ford algorithm

它是一种Distributed backward search algorithm 分布式反向搜索算法

1. For all nodes in the network, do —
   for each neighboring node N, insert the entry (N, link cost to N) in the current routing table T.
2. Do the following procedures for all nodes, while their routing tables may possibly change —

• obtain获得 a neighbor N’s routing table
• For each node Z in the network, calculate: 'new cost = current cost to N + N’s cost to Z’
  – if Z is not in T, insert the pair (N, new cost)
  – else if this new cost is less than the current cost to Z, then replace the entry with (N, new cost)
  没有的话插入，有的话看哪个cheapest替换即可。
  

粗浅的理解来看就是：（不一定对，会修改。）
第一次interation(迭代)是只考虑了与node相邻的点的路径。
第二次迭代的时候，比如算A到F。只考虑中间有一个点的路径。即为A到B到F
第三次迭代的时候，算A到F可以考虑有两个点的路径。比较一下所有的路径的可能性谁的cost最小放在table里。