Notes On Chapter Eight -- LAN Technology And Network Topology

• 8.1 Introduction
  
  
  • This chapter discusses LAN hardware and topology
    
    
  
• 8.2 Direct Point-to-Point Communication
  
  
  • The early "networks" were point-to-point links between computers.
    
    
  • Advantages: The pair of hosts sharing a link could agree on using any type of modem and any communication protocols they liked. Security and privacy were relatively easy to enforce.
    
    
  • Disadvantages: When each pair of computers is connected by a link, the number of links required is proportionate to the square of the number of computers. N hosts ==> N(N-1)/2 links. This is not practical when the number of computers becomes large. Besides, multiple links may follow the same physical path.
    
    
• 8.3 Shared Communication Channels
  
  
  • Inexpensive shared networks were invented in the 60's and 70's so that people could build cost-effective small local networks -- these came to be known as LAN's.
    
    
  • Shared networks don't work well for WAN's. Hosts have to take turns using the network and this requires a lot of communication among the hosts. It is difficult to support high bandwidth communications on long-haul networks, and of course the long distances introduce delay.
    
    
  • point-to-point links incur less communication overhead and work out to be more efficient for long-haul networking.
    
    
• 8.4 Significance Of LANs And Locality Of Reference
  
  
  • In computer networks most traffic is local -- in other words most of the time hosts are communicating with hosts that are nearby. This phenomenon is termed "physical locality of reference."
    
    
  • LAN's are very important because of locality of reference. They support the 'most popular' kind of communication: local communication.
    
    
  • There is also temporal locality of reference. Hosts that have communicated recently are likely to communicate again soon.
    
    
• 8.5 LAN Topologies
  
  
  
  • 8.5.1 Star Topology
    
    
    • all computers attach to a central point (hub).
      
      
    • Advantage: a cut cable affects only one host.
      
      
    • Disadvantage: hub failure brings down the whole network.
      
      
    
  • 8.5.2 Ring Topology
    
    
    • computers are connected in a loop, first to second, second to third, and so on until the last is connected to the first.
      
      
    • Advantage: token ring networks operate efficiently at maximum capacity.
      
      
    • Disadvantage: token rings are vulnerable to link failures. It takes only two link failures to partition any ring.
      
      
    
  • 8.5.3 Bus Topology
    
    
    • all computers on the network share a single link -- the bus -- an Ethernet cable for example.
      
      
    • Hosts have to cooperate to decide which host gets to use the bus next.
      
      
    • Advantage: Only one "wire" is required so the network is inexpensive.
      
      
    • Disadvantage: the whole network may go down if the cable is cut.
      
      
  • 8.5.4 The Reason For Multiple Topologies: each topology has advantages and disadvantages. None is the all-round best for all purposes.
    
    
• 8.6 Example Bus Network: Ethernet
  
  
  • 8.6.1 History Of Ethernet
    
    
    • Invented by Robert Metcalfe at Xerox Corps Palo Alto Research Center in 1973.
      
      
    • Operating Characteristics of the original 10Base5 Ethernet:
      • Coaxial cable - the (luminiferous) ether
      • Cable segments limited to 500 meters -- about 1500 feet.
      • Hosts tap into the cable at a min separation of 3 meters.
      
      
    • There are Ethernets standards for 10, 100, and 1000 Mbps. The 1000 Mbps (1 Gigabit) Ethernet standard was ratified in 1998.
      
      
    • 10 Gigabit Ethernet standards and technology have existed since 2002. A 100 Gigabit standard is under development.
      
      
    • The higher speeds are more geared toward switched LAN's with fiber links. However there are modes in which 100 and 1000 Mbps Ethernet can employ hubs and/or copper wiring.
      
      
    
  • 8.6.2 Ethernet Transmission and Manchester Encoding
    
    
    • Classic Ethernet data is encoded as rising (1-bit) or falling (0-bit) voltages.
      
      
    • The "edge" has to be transmitted in the middle of a time slot.
      
      
    • Changes to the voltage that occur at the boundaries of a time slot do not "count" as data. However such changes must take place in order to send the previous bit value again.
      
      
    • The Manchester scheme is synchronous -- the receiver has to know where the centers and edges of the time slots are.
      
      
    • To synchronize with the receiver, the sender transmits a preamble of 64 alternating 1's and 0's
      
      
    
  • 8.6.3 Sharing On An Ethernet
    
    
    • A sender's electrical signals propagate at about 70% of the speed of light and are soon sensed by all hosts on the segment.
      
      
    • Interference between the signals of two senders would make both signals unintelligible to the receiver.
      
      
    • Only one host sends at a time.
      
      
• 8.7 Carrier Sense On Multi-Access Networks (CSMA)
  
  
  • The mechanism used on an Ethernet to coordinate transmission is called carrier sense multi-access (CSMA). A host wanting to transmit "listens" on the ether and waits until there is no electrical (carrier) activity before beginning to transmit.
    
    
• 8.8 Collision Detection And Backoff With CSMA/CD
  
  
  • Even with CSMA, collisions can occur if two hosts start to send concurrently.
    
    
  • Senders must listen on the ether for sign of a collision (interference). This is called collision detect (CD).
    
    
  • After a collision the hosts each choose a random delay less than some maximum d and attempt to retransmit after that delay. (The value of d used is 102.4 μs - twice the estimated worst-case round trip propagation time.)
    
    
  • If there is another collision immediately after the first, the host chooses its next random delay between 0 and 2d. If collisions continue, the host will continue to double the maximum delay.
    
    
• 8.9 Wireless LANs And CSMA/CA
  
  
  • There are wireless LAN's in which each host transmits on the same frequency. Hosts use a variation on CSMA/CD called carrier sense multi-access with collision avoidance (CSMA/CA).
    
    
  • CSMA/CD won't work properly in a wireless LAN because one host may not be able to detect the transmissions of another host.
    
    For example, 2 may be able to receive transmissions from both 1 and 3, but 1 and 3 may not be in range of each other. If 1 is currently sending a packet to 2, 3 cannot detect the transmission. Unless 3 finds out in some other way, it might send a packet to 2 while 1 is sending its packet to 2. This would create interference in the vicinity of 2.
    
    
  • Using collision avoidance, if 1 wants to send a frame to 2, 1 first sends out a short control message.
    
    
  • If 2 gets the control message, it sends out another control message saying it is ready to receive.
    
    
  • All hosts within range of 2 hear the control message sent out by 2. 1 sends its frame while all other hosts within range of 2 refrain from transmitting.
    
    
  • Control messages can collide, but this is handled with exponential backoff. For example, 1 backs off if it doesn't get the acknowledgment from 2 within a certain time limit.
    
    
• 8.10 Another Example Bus Network: Local Talk
  
  
  • Macintosh Local Talk is a kind of bus network. (quite obsolete now -- supplanted first by EtherTalk and then various other wired and wireless LAN technologies. Apple Talk is now so old that probably the only place you will still find it is in some of California's public schools. :-)
    
    
  • Local Talk uses a version of CSMA/CA
    
    
  • Local Talk has more limited range than Ethernet and much lower bandwidth -- 230.4 Kbps -- only 2.3% of the bandwidth of Ethernet.
    
    
  • Local Talk was cheap and easy to install. It was easy and cheap to put printers on the network that all the hosts could share.
    
    
• 8.11 Ring Topology And Token Passing
  
  
  • Token rings pass bits around the ring from host to host.
    
    
  • Only one host at a time is allowed to transmit -- the host currently in possession of a special bit pattern called the token.
    
    
  • Hosts pass the token around the ring. The host with the token sends one frame and then sends the token.
    
    
  • The sent data gets all the way around the ring. This is how the sender verifies that the frame got through without error.
    
    
  • Passing of the token is done very rapidly by low-level hardware without intervention of the host CPU.
    
    
  • The intended recipient makes a copy of the frame as it passes by.
    
    
• 8.12 Self-Healing Token Passing Networks
  
  
  • Some early ring technologies were vulnerable to link and/or host failure.
    
    
  • Fiber Distributed Data Interlink (FDDI) is a now-obsolescent technology that uses a pair of counter rotating-rings.
    
    
  • There is an automatic "self-healing" effect if a link or host fails.
    
    
• 8.13 Example Star Network: ATM
  
  
  • ATM is an acronym for Asynchronous Transfer Mode
    
    
  • Each hosts connects to an ATM switch which acts as a kind of hub.
    
    
  • When host X needs to send a frame to host Y, the switch connects them electronically.
    
    
  • The data rate for an ATM switch is upwards of 155 Mbps. Typically computers are attached to the switch with optical fiber else that connection might tend to be a bottleneck.
    
    
• 8.14 Summary