Wednesday, November 21, 2012

Notes - Ethernet

The following are notes from Tanenbaum's Computer Networks 5th Edition
  • classic ethernet solves the multiple access problem using the methods studied already
  • switched ethernet, uses switches which connect computers in a different way
Classic Ethernet Physical Layer
  • Ethernet named after luminiferous ether, EM substance necessary for propagation that was disproven
  • DIX standard became IEEE 802.3 standard made by Xerox
  • Thick Ethernet
    • original architecture of ethernet
    • one long ether cable where you attach stations through a transceiver and an interface cable
    • if you reach maximum cable length on the ether, you amplify the signal with a repeater
    • markings every 2.5 meters to show where to attach computers
  • Thin Ethernet
    • connects with industry standard BNC connectors
    • handles smaller number of machines, 30 as opposed to 100
Classic Ethernet MAC Sublayer Protocol
  • formatting used to send frames is shown below
    • Preamble field with the bit pattern 10101010, for all data frames except the last frame where the last two bits are set to 11
      • The last two bits are called the start of frame delimiter
    • the next two fields are the addresses for the source and destination
      • group addresses can be put into this area to allow for multicasting
      • if the address is completely composed of 1s it will become a broadcast
      • station source addresses are globally unique, assigned centrally by IEE
      • any station can uniquely address another
      • first 3 bytes of the address field are used for OUI(Organizationally Unique Identifier)
      • Manufacturers are assigned blocks of 224 addresses, they assign the last three bytes of the address and programs the complete address into the NIC before being sold
    • Type or Length field
      • tells the receiver what to do with the frame
      • IEE 802.3 standard calls it Type, but it also carries the length of the frame
      • LLC(Logical Link Control) conveys the 2 bytes of protocol type information
    • Data field
      • up to 1500 bytes
        • somewhat arbitrarily chosen but based on loosely based on receiver RAM capabilities
        • also minimum frame length
    • Pad field
      • if the data portion is less than 46 bytes, this field fills out to minimum size
      • useful for collision detection, when transceiver detects collision it truncates current frame, so bits and pieces of frame appear on the cable, so it allows detection from garbage data
      • prevents completion of transmission of a short frame before the first bit reaches the far end of a cable
        • i.e. if a frame is very short, before the noise burst of 48 bits that signals that collision has occurred gets to the station, it may have already accepted the packet
      • for 10-Mbps LAN with maximum length of 2500 meters and 4 repeater, 802.3 specification
        • propagation time is roughly 50 microseconds, so shortest frame must be this long at least
    • Checksum field
      • as is stated, it simply sends the checksum for error detection/correction
      • 32 bit CRC
CSMA/CD with Binary Exponential Backoff
  • classic ethernet uses 1-persistent CSMA/CD algorithm
  • however we can use a different random interval algorithm binary exponential backoff
    • after collision, time is divided into discrete slots whose length is equal to the worst-case round-trip propagation time on the ether 2t
    • after first collision each station waits either 0 or 1 slot times at random before retrying
    • if they collide again, each picks 0,1,2, or 3 at random and waits, so after i collision a random number between 0 and 2i-1 is chosen
    • after 10 collisions randomization is frozen at 1023 slots
    • after 16 collisions, failure is reported to computer
Ethernet Performance
  • examine the performance of classic ethernet under heavy and constant loads, k stations ready to transmit
    • k can transmit with probability p, A is the chance to acquire a channel
    • A = kp(1 - p)k-1
    • A is maximized when p = 1/k with A approaching 1/e as k goes to infinity
    • probability that the contention interval has j slots in it is A(1 - A)j-1
    • mean number of slots per contention is
    • since each slot has duration 2t, mean contention interval w is 2t/A
    • assuming optimal p, mean number of contention slots is never more than e so the interval is at most 2te or about 5.4 t
    • so if the mean frame takes P seconds to transmit, the channel efficiency is given by
    • Then we can rework this in terms of network bandwidth B, cable length L, and the speed of signal propogation c. 
    • P = F/B giving us
  • when second term in denominator is large, network efficiency will be low
  • almost all theoretical work assumes poisson traffic, but real data is self-similar or bursty over a range of time scales
Switched Ethernet
  • Ethernet began to evolve from the single long cable architecture to hubs and switches described in the following diagram
  • hubs do not increase capacity because they are logically equivalent to a single long cable, all the wires run to the hub and can only send one at a time
  • switches connects all ports to all other ports, output frames to ports where the frames are destined
  • in a hub all stations are in the same collision domain so must use CSMA/CD algorithm
  • switches have no collisions
  • multiple frames can be sent simultaneously
  • large performance boost
  • LAN interfaces have a promiscuous mode
    • all frames given to each computer
    • preferable to be encrypted but if this feature is there, its a security risk
Fast Ethernet
  • speed of 10 Mbps coming under pressure
  • 802.3 committee in 1992 proposal to come up with a faster LAN
  • basic plan
    • keep all the old frame formats, interfaces and procedural rules
    • reduce bit time form 100 nsec to 10 nsec
    • all fast Ethernet use hubs and switches
    • support for wires to Category 5 twisted pair as a standard
    • Category 3 UTP scheme 100Base-T4 signaling speed of 25 MHz
      • only slightly faster than 20 MHz from ethernet
      • however it requires 4 twisted pairs
      • fell to the wayside because it was complicated and tied up phone lines
    • Category 5 UTP 100Base-TX Ethernet came to dominate market, wires can handle clock rates of 125 MHz
      • 4B/5B encoding is used, 125 MHZ used to provide 100Mbps transmission
      • full duplex
    • 100Base-FX uses two strands of multimode fiber, one in each direction so it can be used in full duplex 100 Mbps in each direction
    • connections all done with hubs or switches, minimum frame size and max cable length must be maintained, so minimum frame size of 64 bytes must go up, or max cable length of 2500 m must come down
      • choice was to decrease distance between hubs by factor of 10
  • All fast ethernet switches can handle both 10 Mbps and 100 Mbps stations, to make upgrading easy, autonegotiation
Gigabit Ethernet
  • 802.3ab in 1999, successor to fast Ethernet
  • same as goals of fast ethernet, increase performance tenfold while maintaining compatibility
  • use point to point links, like fast ethernet, full-duplex and half-duplex mode
  • CSMA/CD protocol is not used, max length of the cable is determined by signal strength issues
  • 10, 100, 1000 Mbps
  • length restriction is extremely painful here, max cable length down to below 200 meters
    • in order to get it up to 200 meters use of carrier extension
    • tells hardware to add its own padding to extend frames to 512 bytes
    • frame bursting transmit a concatenated sequence of multiple frames in a single transmission
    • if the total burst less than 512 bytes, pads it again
  • maintain backwards compatibility
  • short wavelength signaling can be achieved with cheaper LEDs
  • send bits we use 8B/10B encoding
  • requires new copper or fiber cables, 1000Base-T came in to fill this
  • all 4 twisted pairs used, over each wire, 5 voltage levels carry 2 bits used for signaling
  • gigabit Ethernet supports flow control
  • Jumbo Frames, frames allowed longer than 1500 bytes up to 9 KB
10-Gigabit Ethernet
  • 802 committee once again ordered to do same thing improve on gigabit ethernet
  • used in data centers, and high end routers, switches and servers, and enabling entire metropolitan area networks
  • data bit moved up to 64B/66B
  • uses UTP cables, category 6a wiring
  • physical layer is very involved
  • not quite on the market yet, IEEE moved out to standardize 100 Gbps
Retrospective on Ethernet
  • Ethernet has been around for a long time
  • simple and flexible, reliable cheap and easy to maintain
  • interworks with TCP/IP
  • evolve in speed with backwards compatibility
  • FDDI(Fiber Distributed Data Interface), Fibre Channel
    • did not make because not compatible with Ethernet
  • KISS(Keep it simple stupid)
  • Carrier-grade Ethernet is being worked on to let network providers offer services to metropolitan and wide area networks


Back to ToC

No comments:

Post a Comment